The DOI matches its Crossref record; 27 of the 31 references it cites were found in a registry or an index, and 4 are not checked yet; 2 of the 6 sources its claims cite were checked for support.
Safe Reinforcement Learning via Shielding
Open accessAbstract
Reinforcement learning algorithms discover policies that maximize reward, but do not necessarily guarantee safety during learning or execution phases. We introduce a new approach to learn optimal policies while enforcing properties expressed in temporal logic. To this end, given the temporal logic specification that is to be obeyed by the learning system, we propose to synthesize a reactive system called a shield. The shield monitors the actions from the learner and corrects them only if the chosen action causes a violation of the specification. We discuss which requirements a shield must meet to preserve the convergence guarantees of the learner. Finally, we demonstrate the versatility of our approach on several challenging reinforcement learning scenarios.
Verified 2026-10-24 · 4 references not checked yet
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper proposes shielded reinforcement learning, where a reactive shield synthesized from a temporal-logic safety specification and an environment abstraction monitors learner actions and replaces them only when needed to avoid unsafe behavior. It proves correctness and minimal interference, gives conditions for preserving convergence, and demonstrates safety and often faster learning in four simulated domains.
Key findings
- Shields can enforce safety during both learning and execution without changing the reward signal, and they are minimally interfering: they block an action only when it could lead to a specification violation under the abstraction.
- In the water-tank example, the product game had 602 states, and the shield prevented opening the inflow valve above level 93 and forced opening when the level became too low.
- Shield synthesis was fast in the reported experiments: under 2 seconds for the grid-world cases, about 2 seconds for the self-driving car, under 1 second for the water tank, and under 7 seconds for Pacman.
- In the 15x9 grid world, only the shielded setting with a ranking of 3 actions and no penalty found the optimal path; in the water-tank experiments, shielded Q-learning and SARSA reached optimal policies in significantly fewer episodes than unshielded versions; in the self-driving car case, the shielded learner never crashed and learned faster; in Pacman, the shielded agent cleared the game from the first episode.
Methods
- The approach specifies safety in linear temporal logic, converts it to a deterministic safety automaton, builds a conservative finite-state abstraction of the MDP environment, and solves a safety game to synthesize a shield.
- The shield receives the learner’s action or ranked list of actions and executes the highest-ranked safe action; it deviates only if every ranked action is unsafe, in which case it chooses some safe action.
- The paper proves correctness and minimal interference, constructs a product MDP to argue convergence for standard MDP learners, and evaluates the method in simulations: 9x9 and 15x9 grid worlds, a 480x480-pixel self-driving car task, the water-tank controller, and an 18x5 Pacman-like maze, using tabular Q-learning, deep Q-networks, and approximate Q-learning.
Limitations
- The paper states that the main downside is the need for an approximate model or abstraction indicating when actions are unsafe, and argues this is unavoidable when allowed actions depend on the environment state.
- The convergence argument requires that when the learner’s ranking contains no safe action, the shield’s choice among safe actions follows a fixed probability distribution over time, and it assumes the learner can observe the shield state.
- Evident from the text: the evaluation is entirely simulation-based, with no real-world deployment; the approach relies on the abstraction being conservative, and the shield must remain active after learning because unsafe actions may still be part of the learned policy.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
2 of 6 cited sources checked · 0 supported · 1 flagged · accuracy not yet measured
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The results show that only the unshielded versions experience negative rewards.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Fig. 11 shows that both shielded (dashed lines) and unshielded Q-learning and SARSA experiments (solid lines) do reach an optimal policy.”
§ Safe Reinforcement Learning via Shielding > Experiments
Strength:plain· not yet calibratedSingle caseTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In all of them, the learning performance of the shielded agents improved compared to the unshielded case.”
§ Safe Reinforcement Learning via Shielding > Conclusion
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As expected, in the shielded case, pacman is always able to clear the game starting from the first episode.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, the shielded implementations reach the optimal policy in a significantly shorter time than the unshielded implementations.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the computational cost of the construction of the shield depends on the complexity of the specification and a very abstract version of the system”
§ Safe Reinforcement Learning via Shielding > Safety in Formal Methods.
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesA conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“since the shield corrects unsafe actions in the learning and deployment phases.”
§ Safe Reinforcement Learning via Shielding > Related Work Safety in RL.
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Even though the inner working of learning algorithms is often complex, shielding with respect to critical safety specifications may be manageable (as we demonstrate in upcoming sections).”
§ Safe Reinforcement Learning via Shielding > Environment
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes? Based: unclearA conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, to reason about when exactly a specification violation cannot be avoided, we have to give a (coarse finite-state) abstraction of the environment dynamics.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesA conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We introduce a new approach to learn optimal policies while enforcing properties expressed in temporal logic.”
§ Safe Reinforcement Learning via Shielding > Abstract
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passes
- Probably not a substantial claim. 17
- It may state two separate claims as one. 17
At least one cited source doesn’t support this claim or points to a different paper, and none confirmed it — see the verdicts below.
In the paper
“An exploration process is called safe if no undesirable states are ever visited, which can only be achieved through the incorporation of external knowledge (García and Fernández 2015;Moldovan and Abbeel 2012).”
§ Safe Reinforcement Learning via Shielding > Related Work Safety in RL.
What the sources say
- Not checked — no full text of the sourcecites García and Fernández 2015
- Disputedcites Moldovan and Abbeel 2012 sourceExperimental — accuracy not yet measured
The Safety Constraint section (p. 3) defines safety as preserving ergodicity with a user-specified probability, not as never visiting an undesirable state. The Safety Counter-Examples section (p. 3) explicitly argues against defining safety by safe and unsafe states. The paper uses a prior belief and information learned during exploration, but does not say that incorporating external knowledge is the only way to achieve safety. (LLM, conf=0.82)
Checked by a reasoning model — may be wrong.
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✓ Based: passesProbably not a substantial claim. The paper’s conclusions may not depend on it.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Unlike the approach in the paper, their technique does not work with function approximation, which prevents it from being used in complex scenarios.”
§ Safe Reinforcement Learning via Shielding > Safety in Formal Methods.
What the sources say
- Partly supportedcites Wen, Ehlers, and Topcu 2015 sourceExperimental — accuracy not yet measured
Section V (Examples; page unavailable) shows that learning time and iterations grow as the grid grows, and describes a trade-off between performance and computation cost. This supports a scalability concern, but the paper does not say that its technique cannot work with function approximation or that it cannot be used in complex scenarios. (LLM, conf=0.76)
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
- Partly supported
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The plot in Fig. 10 shows that the accumulated rewards for unshielded RL (red, dashed) increases over time, but the car still crashes at the end of the simulation.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In scenarios with |rank t | = 1 (red) or with penalties (solid), the agent computes a suboptimal path.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Only the shielded version with |rank t | = 3 without penalty (blue, dashed) finds the optimal path, resulting in a higher average reward.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Furthermore, the shielded version is not only safe, but also tends to learn more rapidly.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The shielded version without punishment (blue, solid) learns more rapidly than the unshielded learner and never crashes.”
§ Safe Reinforcement Learning via Shielding > Experiments
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
In the paper
“For specifications in LTL, it is known how to check if it is a safety language and how to compute a safety automaton that represents it (Kupferman and Vardi 2001).”
§ Safe Reinforcement Learning via Shielding > The set of all distributions on X is denoted by Distr(X).
What the sources say
- Not checked — no full text of the sourcecites Kupferman and Vardi 2001 source
Load-bearingStrength:boosted· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“An abstraction of an MDP describes how its executions can possibly evolve, and provides the needed information about the environment to allow planning ahead w.r.t. the safety properties of interest.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✗ Checkable: fails✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and is independent of the state space components of the system to be controlled that are irrelevant for enforcing the safety specification.”
§ Safe Reinforcement Learning via Shielding > Safety in Formal Methods.
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It ensures safety and minimum interference.”
§ Safe Reinforcement Learning via Shielding > Environment
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The shield is computed upfront from the safety part of the given system specification and an abstraction of the agent's environment dynamics.”
§ Safe Reinforcement Learning via Shielding > Environment
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This boundary helps to separate the concerns, e.g., safety and correctness on one side and convergence and optimality on the other”
§ Safe Reinforcement Learning via Shielding > The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The latter is necessary as the valve has to stay open for at least three time steps.So as the inflow may be up to 2 liters/second during this time and the outflow may be 0, there is otherwise an overflow risk.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This approximation may have finitely many states even if the MDP has infinitely many states and/or is only approximately known.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Last but not least, the shielding framework is compatible with mechanisms such as function approximation, employed by learning algorithms in order to improve their scalability.”
§ Safe Reinforcement Learning via Shielding > The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)
Strength:plain· not yet calibratedCausal:correlational· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✗ Based: failsProbably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The algorithms we present for the computation of shields make relatively mild assumptions on the input-output structure of the learning algorithm (rather than its inner working).”
§ Safe Reinforcement Learning via Shielding > Environment
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Reactive synthesis enforces ϕ s by solving a safety game built from ϕ s and an abstraction of the environment in which the policy is to be executed.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Planning ahead does not require the environment dynamics to be completely known in advance.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“All of the above approaches have in common that a faithful, yet precise enough, abstraction of the physical environment is required, which is not only difficult to obtain in practice, but also introduces the mentioned computa-tional burden.”
§ Safe Reinforcement Learning via Shielding > Safety in Formal Methods.
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“and provides a basis for the convergence analysis of a shielded RL algorithm.”
§ Safe Reinforcement Learning via Shielding > The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This ensures that paths that model traces that cannot occur in M are rejected by ϕ M as early as possible.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✗ Load-bearing: fails✗ Based: failsProbably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As the shield is generated using the specification, it plans ahead for this not to happen, so it must prevent the opening of the inflow valve if the level is above 93.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our setup introduces a clear boundary between the learning agent and the shield.”
§ Safe Reinforcement Learning via Shielding > The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18)
Load-bearingStrength:boosted· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Given that the environment is often represented as an MDP, such an abstraction has to be conservative w.r.t. the behavior of the real MDP.”
§ Safe Reinforcement Learning via Shielding > Safety, Abstractions, and Games
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✗ Based: failsProbably not a substantial claim. The paper may give no reason why it holds.
- Rests on an assumption: The environment is often represented as an MDP.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Control methods based on RL partly address this problem, but do not typically incorporate any correctness guarantees.”
§ Safe Reinforcement Learning via Shielding > Safety in Formal Methods.
Load-bearingStrength:plain· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Shielding exploits this possibility.”
§ Safe Reinforcement Learning via Shielding > Conclusion
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesProbably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Since an unsafe action is always mapped to a safe one, this does not pose a problem and the agent never has to learn to avoid unsafe actions.Consequently, the shield is (again) needed during the learning and execution phases.”
§ Safe Reinforcement Learning via Shielding > Shielded Reinforcement Learning
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our experiments show, however, that in applications in which safe learning is needed, the effort to construct an abstraction is well-spent, as our approach not only makes learning safe, but also shows great promise of improving learning performance.”
§ Safe Reinforcement Learning via Shielding > Conclusion
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The agent assigns a punishment r t+1 < 0 to a 1 t and learns that selecting a 1 t at state s t is unsafe, without ever violating ϕ s .”
§ Safe Reinforcement Learning via Shielding > Shielded Reinforcement Learning
Strength:plain· not yet calibratedCausal:direct· not yet calibratedTests✓ Asserted: passes✓ Checkable: passes? Load-bearing: unclear✓ Based: passesIt may state two separate claims as one.
- The authors state this as an assumption: The environment can be modeled as an MDP M = (S, s_I, A, P, R).
- The authors state this as an assumption: We have constructed an abstraction ϕ_M.
- The authors state this as an assumption: The learner accepts elements from S × Q as state input (for the state space of the shield Q).
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, there is no guarantee that unsafe actions are not part of the final policy.Therefore, the shield has to remain active even after the learning phase.”
§ Safe Reinforcement Learning via Shielding > Shielded Reinforcement Learning
Tests✓ Asserted: passes✓ Checkable: passes✓ Load-bearing: passes✓ Based: passesIt may state two separate claims as one.
The authors state this as an assumption.
The specification ϕ is a set of properties {ϕ1,...,ϕl} such that L(ϕ) is the intersection of the languages of its properties.
“We assume ϕ is a set of properties {ϕ 1 , . . ., ϕ l } such that L(ϕ) = i L(ϕ i ).”
The paper relies on this without support.
The environment is often represented as an MDP.
“Given that the environment is often represented as an MDP, such an abstraction has to be conservative w.r.t. the behavior of the real MDP.”
The paper relies on this without support.
Safety or more generally correctness during learning and execution of controllers has attracted significantly less attention than convergence, optimality and data-efficiency of learning algorithms.
“While convergence, optimality and data-efficiency of learning algorithms are relatively well understood, safety or more generally correctness during learning and execution of controllers has attracted significantly less attention.”
The authors state this as an assumption.
The environment can be modeled as an MDP M = (S, s_I, A, P, R).
“Assumption 1 (i) The environment can be modeled as an MDP M = (S, s I , A, P, R).”
The authors state this as an assumption.
We have constructed an abstraction ϕ_M.
“(ii) We have constructed an abstraction ϕ M .”
The authors state this as an assumption.
The learner accepts elements from S × Q as state input (for the state space of the shield Q).
“(iii) The learner accepts elements from S × Q as state input (for the state space of the shield Q).”
The authors state this as an assumption.
Some abstraction ϕ_M = (Q_M, q_0,M, A × L, δ_M, F_M) of M for some MDP observer function f : S → L is given.
“We assume some abstraction ϕ M = (Q M , q 0,M , A × L, δ M , F M ) of M for some MDP observer function f : S → L to be given.”
The authors state this as an assumption.
ϕ_s has Σ = L × A, i.e., it reads the part of the system behavior that the abstraction is concerned with.
“Since ϕ s models a restriction of the traces of the MDP and the learner together that we want to enforce, we assume it to have Σ = L × A, i.e., it reads the part of the system behavior that the abstraction is concerned with.”
The authors state this as an assumption.
A shield S = (Q_S, q_0,S, Σ_I,S, Σ_O,S, δ_S, λ_S) is given for M and for some MDP labeling function f : S → L.
“Assume also that a shield S = (Q S , q 0,S , Σ I,S , Σ O,S , δ S , λ S ) is given for M and for some MDP labeling function f : S → L.”
The authors state this as an assumption.
Whenever an action ranking is chosen by the learner that does not contain a safe action, there is a fixed probability distribution over the safe actions executed instead, constant over time.
“Note that this argument requires that whenever an action ranking is chosen by the learner that does not contain a safe action, there is a fixed probability distribution over the safe actions executed instead.”
The authors state this as an assumption.
The learner has access to the state of the shield and can base its actions on it.
“Note that we made use of the fact that the learner has access to the state of the shield and can base its actions on it in this argument.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“To provide system-level, physically absolute safety guarantees, SafeCodeRL incorporates a hard shielding mechanism [37].”
From SafeCodeRL: Security-Constrained Multi-Agent Reinforcement Learning for Trustworthy LLM-Generated IoT/CPS Software (2026) · cites this paper as [37]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Abstraction alphabet order conflicts with its transitions
MinorThe abstraction is declared to read action-label pairs, A × L, but the product-game transition supplies it label-action pairs, (l, a). These are different input orders for the same abstraction automaton.
“ϕ M = (Q M , q 0,M , A × L, δ M , F M )”
“(δ(q, (l, a)), δ M (q, (l, a))), for all (q, q M ) ∈ G, l ∈ L, a ∈ A”
Safe specification states are called error states
MinorThe game definition includes states in F × Q M among its safe states, but the example says those same states are merged into an error state. Taken literally, that reverses which specification states represent an error in the reported game construction.
“if we merge all states in F × Q M into a single error state”
“F g = (F × Q M ) ∪ (Q × (Q M \ F M )).”
Citations
Cites
Human-level control through deep reinforcement learning
Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A. A., Veness, J., Bellemare, M. G., Graves, A., Riedmiller, M., Fidjeland, A. K., Ostrovski, G., et al. (2015) · Nature · doi:10.1038/nature14236 · Cited by 30,357
Reinforcement Learning: An Introduction
Sutton, R. S., & Barto, A. G. (1998) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.1998.712192 · Cited by 27,225
Reinforcement Learning: An Introduction
Sutton, R. S., & Barto, A. (2005) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.2004.842673 · Cited by 25,268
(2008) · Cited by 4,811
Handbook of theoretical computer science
Leeuwen, J. V. (1990) · Medical Entomology and Zoology · Cited by 4,259
Valiant, L. G. (1984) · ACM eBooks · doi:10.1145/3749740.3749746 · Cited by 4,211
Valiant, L. G. (1984) · Communications of the ACM · doi:10.1145/1968.1972 · Cited by 3,279
Emerson, E. (1991) · Handbook of Theoretical Computer Science, Volume B: Formal Models and Sematics · doi:10.1016/b978-0-444-88074-1.50021-4 · Cited by 3,268
The Arcade Learning Environment: An Evaluation Platform for General Agents
Bellemare, M. G., Naddaf, Y., Veness, J., & Bowling, M. (2013) · Journal of Artificial Intelligence Research · doi:10.1613/jair.3912 · Cited by 2,161
The temporal logic of programs
Pnueli, A. (1977) · 18th Annual Symposium on Foundations of Computer Science (sfcs 1977) · doi:10.1109/sfcs.1977.32 · Cited by 1,523
(2015) · Journal of Machine Learning Research · Cited by 1,119
Principles of Model Checking (Representation and Mind Series)
Christel Baier, A., & Katoen, J.-P. (2008) · The MIT Press eBooks · Cited by 1,021
Temporal Verification of Reactive Systems: Safety
Manna, Z., & Pnueli, A. (1995) · Medical Entomology and Zoology · Cited by 947
Model Checking of Safety Properties
Kupferman, O., & Vardi, M. Y. (2001) · Formal Methods in System Design · doi:10.1023/a:1011254632723 · Cited by 566
Teachable robots: Understanding human teaching behavior to build more effective robot learners
Thomaz, A. L., & Breazeal, C. (2007) · Artificial Intelligence · doi:10.1016/j.artint.2007.09.009 · Cited by 414
Receding Horizon Temporal Logic Planning
Wongpiromsarn, T., Topcu, U., & Murray, R. M. (2012) · IEEE Transactions on Automatic Control · doi:10.1109/tac.2012.2195811 · Cited by 292
Self-citationThomaz, A. L., & Breazeal, C. (2006) · Cited by 259
Model Checking of Safety Properties
Kupferman, O., & Vardi, M. (1999) · Lecture notes in computer science · doi:10.1007/3-540-48683-6_17 · Cited by 160
Safe Exploration in Markov Decision Processes
Moldovan, T. M., & Abbeel, P. (2012) · arXiv · doi:10.48550/arxiv.1205.4810 · Cited by 152
Discrete-time control for rectangular hybrid automata
Henzinger, T. A., & Kopke, P. W. (1999) · Theoretical Computer Science · doi:10.1016/s0304-3975(99)00038-9 · Cited by 127
Safety-Constrained Reinforcement Learning for MDPs
Junges, S., Jansen, N., Dehnert, C., et al. (2016) · Lecture notes in computer science · doi:10.1007/978-3-662-49674-9_8 · Cited by 93
Self-citationSafe Exploration Techniques for Reinforcement Learning – An Overview
Pecka, M., & Svoboda, T. (2014) · Lecture notes in computer science · doi:10.1007/978-3-319-13823-7_31 · Cited by 83
Correct-by-synthesis reinforcement learning with temporal logic constraints
Wen, M., Ehlers, R., & Topcu, U. (2015) · 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros.2015.7354078 · Cited by 59
Self-citationSafety first: a two-stage algorithm for the synthesis of reactive systems
Sohail, S., & Somenzi, F. (2012) · International Journal on Software Tools for Technology Transfer · doi:10.1007/s10009-012-0224-3 · Cited by 32
Shield Synthesis: Runtime Enforcement for Reactive SystemsPreprint
Bloem, R., Könighofer, B., Koenighofer, R., et al. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1501.02573 · Cited by 30
Self-citationSynthesis of Shared Autonomy Policies With Temporal Logic Specifications
Fu, J., & Topcu, U. (2015) · IEEE Transactions on Automation Science and Engineering · doi:10.1109/tase.2015.2499164 · Cited by 24
Self-citationDeCastro, J., & Kress‐Gazit, H. (2016) · International Conference on Hybrid Systems: computation and control · doi:10.1145/2883817.2883832 · Cited by 13
Learning on real robots from experience and simple user feedback
Vidal, P. Q., Iglesias, R., González, M. Á. R., et al. (2013) · Journal of Physical Agents (JoPha) · doi:10.14198/jopha.2013.7.1.08 · Cited by 12
On integrating apprentice learning and reinforcement learning TITLE2
Clouse, J. A. (1997) · Cited by 6
Bloem, R., Könighofer, B., Könighofer, R., et al. (2015) · Lecture Notes in Computer Science · doi:10.1007/978-3-662-46681-0_51
Self-citationTemporal Verification of Reactive Systems
Manna, Z., & Pnueli, A. (1995) · doi:10.1007/978-1-4612-4222-2
Flagged first, then by citations
Cited by
Sorted by citations
Cheng, R., Orosz, G., Murray, R. M., & Burdick, J. W. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33013387 · Cited by 611
“The concept is akin to shielded RL (Alshiekh et al. 2017; Fisac et al. 2018), since the CBF controller compensates for the RL controller to ensure safety, but does not guide exploration of the RL algorithm.”
BackgroundNavigating Occluded Intersections with Autonomous Vehicles Using Deep Reinforcement Learning
Isele, D., Rahimi, R., Cosgun, A., Subramanian, K., & Fujimura, K. (2018) · 2018 IEEE International Conference on Robotics and Automation (ICRA) · doi:10.1109/icra.2018.8461233 · Cited by 379
Ganesh, A. H., & Xu, B. (2021) · Renewable and Sustainable Energy Reviews · doi:10.1016/j.rser.2021.111833 · Cited by 295
Deep Multi-Agent Reinforcement Learning for Highway On-Ramp Merging in Mixed Traffic
Chen, D., Hajidavalloo, M. R., Li, Z., Chen, K., Wang, Y., Jiang, L., & Wang, Y. (2023) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2023.3285442 · Cited by 290
Tran, H.-D., Yang, X.-D., Lopez, D. M., Musau, P., Nguyen, L. V., Xiang, W.-M., Bak, S., & Johnson, T. T. (2020) · International Conference on Computer Aided Verification · doi:10.1007/978-3-030-53288-8_1 · Cited by 306
Metadata
- Authors
- Alshiekh, M., Bloem, R., Ehlers, R., Könighofer, B., Niekum, S., & Topcu, U.
- Affiliations
- The University of Texas at Austin (US)
- Graz University of Technology (AT)
- University of Bremen (DE)
- Venue
- Proceedings of the AAAI Conference on Artificial Intelligence
- Published
- 2018-04-29
- Volume
- 32
- Issue
- 1
- ISSN
- 2159-5399 (ISSN-L)
- 2374-3468
- OpenAlex
- W2751422670
- Cited by
- 37
- References
- 19
- Citations by year
- 2026: 1 · 2025: 1 · 2024: 3 · 2023: 7 · 2022: 3 · 2021: 7 · 2020: 7 · 2019: 4 · 2018: 4
- FWCI
- 2.24
- Type
- conference paper
- Language
- en
- Open access
- diamond
- doi.org
- Locations
- Proceedings of the AAAI Conference on Artificial Intelligence (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Science Foundation (1652113, 1617639, 1646522)
- Austrian Science Fund (S11406-N23, W1255)
- Universität Bremen
- Defense Advanced Research Projects Agency (W911NF-16-1-0001, FA8650-15C)
- Army Research Office (W911NF-15-1-0592, W911NF-15-1-, W911NF-16-1-, W911NF, W911NF-15)
- Air Force Research Laboratory (FA8650-15-C-2546, FA8650-15-, 8650-16-C-2610)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Computer Science › Computational Theory and Mathematics › Formal Methods in Verification
- Physical Sciences › Computer Science › Artificial Intelligence › Logic, Reasoning, and Knowledge
- Keywords
- safe reinforcement learning, temporal logic specifications, reactive systems, convergence guarantees, policy learning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-24, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-24 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
