Exploration in Deep Reinforcement Learning: From Single-Agent to Multiagent Domain
Open accessThe DOI matches its Crossref record; 44 of the 44 references it cites were found in a registry or an index.
Abstract
Deep reinforcement learning (DRL) and deep multiagent reinforcement learning (MARL) have achieved significant success across a wide range of domains, including game artificial intelligence (AI), autonomous vehicles, and robotics. However, DRL and deep MARL agents are widely known to be sample inefficient that millions of interactions are usually needed even for relatively simple problem settings, thus preventing the wide application and deployment in real-industry scenarios. One bottleneck challenge behind is the well-known exploration problem, i.e., how efficiently exploring the environment and collecting informative experiences that could benefit policy learning toward the optimal ones. This problem becomes more challenging in complex environments with sparse rewards, noisy distractions, long horizons, and nonstationary co-learners. In this article, we conduct a comprehensive survey on existing exploration methods for both single-agent RL and multiagent RL. We start the survey by identifying several key challenges to efficient exploration. Then, we provide a systematic survey of existing approaches by classifying them into two major categories: uncertainty-oriented exploration and intrinsic motivation-oriented exploration. Beyond the above two main branches, we also include other notable exploration methods with different ideas and techniques. In addition to algorithmic analysis, we provide a comprehensive and unified empirical comparison of different exploration methods for DRL on a set of commonly used benchmarks. According to our algorithmic and empirical investigation, we finally summarize the open problems of exploration in DRL and deep MARL and point out a few future directions.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
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 · cited 1×
“1) Value-Based Methods: Deep Q -network (DQN) [25] is the most representative algorithm in DRL that derived from Q -learning.”
MethodIntrinsic and Extrinsic Motivations: Classic Definitions and New Directions
Ryan, R. M., & Deci, E. L. (2000) · Contemporary Educational Psychology · doi:10.1006/ceps.1999.1020 · Cited by 18,666 · cited 3×
“Intrinsic motivation is originated from humans inherent tendency to interact with the world in an attempt to have an effect, and to feel a sense of accomplishment [26], [27].”
BackgroundInfluentialMastering the game of Go with deep neural networks and tree search
Silver, D., Huang, A., Maddison, C. J., Guez, A., Sifre, L., van den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M., et al. (2016) · Nature · doi:10.1038/nature16961 · Cited by 15,612
Continuous control with deep reinforcement learning
Lillicrap, T. P., Hunt, J. J., Pritzel, A., Heess, N., Erez, T., Tassa, Y., Silver, D., & Wierstra, D. (2015) · arXiv · doi:10.48550/arxiv.1509.02971 · Cited by 10,782
Proximal Policy Optimization Algorithms
Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017) · arXiv · doi:10.48550/arxiv.1707.06347 · Cited by 10,625 · cited 1×
“Experiments on Montezuma’s Revenge show that ADM successfully extracts the location of the character and achieves a high score of 11618 combined with PPO [108].”
MethodMastering the game of Go without human knowledge
Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A., Guez, A., Hubert, T., Baker, L., Lai, M., Bolton, A., et al. (2017) · Nature · doi:10.1038/nature24270 · Cited by 9,084
Asynchronous Methods for Deep Reinforcement Learning
Mnih, V., Badia, A. P., Mirza, M., Graves, A., Lillicrap, T. P., Harley, T., Silver, D., & Kavukcuoglu, K. (2016) · arXiv · doi:10.48550/arxiv.1602.01783 · Cited by 7,132 · cited 1×
“…RL methods add randomness in action selection based on Q -values, entropy regularization is widely used to promote exploration for RL algorithms [35] with stochastic policies, by adding the policy entropy H (π( a | s )) to the objective function as a regularization term, encouraging the policy…”
MethodDropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y., & Ghahramani, Z. (2015) · arXiv · doi:10.48550/arxiv.1506.02142 · Cited by 6,471
Finite-time Analysis of the Multiarmed Bandit Problem
Auer, P., Cesa-Bianchi, N., & Fischer, P. (2002) · Machine Learning · doi:10.1023/a:1013689704352 · Cited by 6,033
Deep Reinforcement Learning: A Brief Survey
Arulkumaran, K., Deisenroth, M. P., Brundage, M., & Bharath, A. A. (2017) · IEEE Signal Processing Magazine · doi:10.1109/msp.2017.2743240 · Cited by 4,733 · cited 1×
“A few prior papers contain the investigation on exploration methods for DRL [8], [9], [10], [11], [12] and MARL [13], [14], [15].”
MethodA Brief Survey of Deep Reinforcement Learning
Arulkumaran, K., Deisenroth, M. P., Brundage, M., & Bharath, A. A. (2017) · arXiv · doi:10.48550/arxiv.1708.05866 · Cited by 4,733
Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2016) · arXiv · doi:10.48550/arxiv.1612.01474 · Cited by 3,666
Addressing Function Approximation Error in Actor-Critic Methods
Fujimoto, S., van Hoof, H., & Meger, D. (2018) · arXiv · doi:10.48550/arxiv.1802.09477 · Cited by 3,650 · cited 1×
“The stochastic process can be selected to suit the environment, e.g., an Ornstein–Uhlenbeck process is preferred in physical control problems [33]; more generally, a standard Gaussian distribution is simply adopted in many works [36].”
BackgroundGrandmaster level in StarCraft II using multi-agent reinforcement learning
Vinyals, O., Babuschkin, I., Czarnecki, W. M., Mathieu, M., Dudzik, A., Chung, J., Choi, D. H., Powell, R., Ewalds, T., Georgiev, P., et al. (2019) · Nature · doi:10.1038/s41586-019-1724-z · Cited by 3,455
Self-citationMulti-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
Lowe, R., Wu, Y., Tamar, A., Harb, J., Abbeel, P., & Mordatch, I. (2017) · arXiv · doi:10.48550/arxiv.1706.02275 · Cited by 3,377
Dueling Network Architectures for Deep Reinforcement Learning
Wang, Z., Schaul, T., Hessel, M., van Hasselt, H., Lanctot, M., & de Freitas, N. (2015) · arXiv · doi:10.48550/arxiv.1511.06581 · Cited by 3,300
Self-citationTHOMPSON, W. R. (1933) · Biometrika · doi:10.1093/biomet/25.3-4.285 · Cited by 2,712
Asymptotically efficient adaptive allocation rules
Lai, T. L., & Robbins, H. (1985) · Advances in Applied Mathematics · doi:10.1016/0196-8858(85)90002-8 · Cited by 2,406
A Comprehensive Survey of Multiagent Reinforcement Learning
Busoniu, L., Babuska, R., & De Schutter, B. (2008) · IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews) · doi:10.1109/tsmcc.2007.913919 · Cited by 2,235 · cited 1×
“A few prior papers contain the investigation on exploration methods for DRL [8], [9], [10], [11], [12] and MARL [13], [14], [15].”
MethodCounterfactual Multi-Agent Policy Gradients
Foerster, J., Farquhar, G., Afouras, T., Nardelli, N., & Whiteson, S. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11794 · Cited by 1,873
Curiosity-Driven Exploration by Self-Supervised Prediction
Pathak, D., Agrawal, P., Efros, A. A., & Darrell, T. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw.2017.70 · Cited by 1,815 · cited 4×
“Go [1], Atari [2], StarCraft [3], robotics [4], and other control problems [5], [6], [7].”
BackgroundMethodInfluentialIMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
Espeholt, L., Soyer, H., Munos, R., Simonyan, K., Mnih, V., Ward, T., Doron, Y., Firoiu, V., Harley, T., Dunning, I., et al. (2018) · arXiv · doi:10.48550/arxiv.1802.01561 · Cited by 1,708
Learning to Communicate with Deep Multi-Agent Reinforcement Learning
Foerster, J. N., Assael, Y. M., de Freitas, N., & Whiteson, S. (2016) · arXiv · doi:10.48550/arxiv.1605.06676 · Cited by 1,217
Distributed Prioritized Experience Replay
Horgan, D., Quan, J., Budden, D., Barth-Maron, G., Hessel, M., van Hasselt, H., & Silver, D. (2018) · arXiv · doi:10.48550/arxiv.1803.00933 · Cited by 963
Simple statistical gradient-following algorithms for connectionist reinforcement learning
Williams, R. J. (1992) · Machine Learning · doi:10.1023/a:1022672621406 · Cited by 817
Deep Reinforcement Learning: An Overview
Li, Y. (2017) · arXiv · doi:10.48550/arxiv.1701.07274 · Cited by 762 · cited 1×
“A few prior papers contain the investigation on exploration methods for DRL [8], [9], [10], [11], [12] and MARL [13], [14], [15].”
MethodDistributional Reinforcement Learning With Quantile Regression
Dabney, W., Rowland, M., Bellemare, M., & Munos, R. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11791 · Cited by 612
Exploration in deep reinforcement learning: A survey
Ladosz, P., Weng, L., Kim, M., & Oh, H. (2022) · Information Fusion · doi:10.1016/j.inffus.2022.03.003 · Cited by 525 · cited 1×
“Work [21] is a concurrent work.”
Diversity is All You Need: Learning Skills without a Reward Function
Eysenbach, B., Gupta, A., Ibarz, J., & Levine, S. (2018) · arXiv · doi:10.48550/arxiv.1802.06070 · Cited by 480 · cited 1×
“DIAYN [223] trains the policy to maximize I(S;Z) while minimizing I(A;Z|S) additionally, which pushes the skills away from each other to learn distinguishable skills.”
BackgroundLarge-Scale Study of Curiosity-Driven Learning
Burda, Y., Edwards, H., Pathak, D., Storkey, A., Darrell, T., & Efros, A. A. (2018) · arXiv · doi:10.48550/arxiv.1808.04355 · Cited by 401 · cited 2×
“Existing methods mainly focus on learning state representation [5], [46], [52], [53] to improve robustness when faced with the white-noise problem.”
BackgroundMethodCount-Based Exploration with Neural Density Models
Ostrovski, G., Bellemare, M. G., Oord, A. V. D., & Munos, R. (2017) · arXiv · doi:10.48550/arxiv.1703.01310 · Cited by 377 · cited 2×
“Intrinsic motivation-oriented methods often need additional auxiliary models such as forward dynamics [105] and density model [87] to measure the novelty of states or transitions.”
BackgroundMachado, M. C., Bellemare, M. G., Talvitie, E., Veness, J., Hausknecht, M., & Bowling, M. (2018) · Journal of Artificial Intelligence Research · doi:10.1613/jair.5699 · Cited by 359
Ecoffet, A., Huizinga, J., Lehman, J., Stanley, K. O., & Clune, J. (2021) · Nature · doi:10.1038/s41586-020-03157-9 · Cited by 330
Learning Dexterous In-Hand Manipulation
OpenAI, Andrychowicz, M., Baker, B., Chociej, M., Jozefowicz, R., McGrew, B., Pachocki, J., Petron, A., Plappert, M., Powell, G., et al. (2018) · arXiv · doi:10.48550/arxiv.1808.00177 · Cited by 248
Go-Explore: a New Approach for Hard-Exploration Problems
Ecoffet, A., Huizinga, J., Lehman, J., Stanley, K. O., & Clune, J. (2019) · arXiv · doi:10.48550/arxiv.1901.10995 · Cited by 205 · cited 4×
“…over 9.3k scores by taking advantage of a two-scale novelty mechanism and an advanced distributed architecture base (i.e., R2D2 [113]); Go-Explore [135], [136] achieves superhuman performance through imitating superior trajectory experiences collected in a return-and-explore fashion with access…”
BackgroundMethodInfluentialNever Give Up: Learning Directed Exploration Strategies
Badia, A. P., Sprechmann, P., Vitvitskyi, A., Guo, D., Piot, B., Kapturowski, S., Tieleman, O., Arjovsky, M., Pritzel, A., Bolt, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2002.06038 · Cited by 170
A multi-objective deep reinforcement learning framework
Nguyen, T. T., Nguyen, N. D., Vamplew, P., Nahavandi, S., Dazeley, R., & Lim, C. P. (2020) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2020.103915 · Cited by 152
Intrinsic Motivation and Automatic Curricula via Asymmetric Self-Play
Sukhbaatar, S., Lin, Z., Kostrikov, I., Synnaeve, G., Szlam, A., & Fergus, R. (2017) · arXiv · doi:10.48550/arxiv.1703.05407 · Cited by 149 · cited 1×
“Competitive experience replay [215] introduces self-play [216] between two agents and generates a curriculum”
BackgroundSurprise-Based Intrinsic Motivation for Deep Reinforcement Learning
Achiam, J., & Sastry, S. (2017) · arXiv · doi:10.48550/arxiv.1703.01732 · Cited by 139 · cited 1×
“Later, Achiam and Sastry [104] proposed a more efficient way than VIME to learn the dynamics model, by replacing BNNs with a neural network followed by fully factored Gaussian distributions.”
MethodPlanning to Explore via Self-Supervised World Models
Sekar, R., Rybkin, O., Daniilidis, K., Abbeel, P., Hafner, D., & Pathak, D. (2020) · arXiv · doi:10.48550/arxiv.2005.05960 · Cited by 129 · cited 1×
“Planning to explore [198] also uses ensemble dynamics for uncertainty estimation, and”
BackgroundEX2: Exploration with Exemplar Models for Deep Reinforcement Learning
Fu, J., Co-Reyes, J. D., & Levine, S. (2017) · arXiv · doi:10.48550/arxiv.1703.01260 · Cited by 95
Thananjeyan, B., Balakrishna, A., Rosolia, U., Li, F., McAllister, R., Gonzalez, J. E., Levine, S., Borrelli, F., & Goldberg, K. (2020) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2020.2976272 · Cited by 83 · cited 1×
“From another angle, some works make use of given demonstrations and devise safe exploration methods, e.g., by fitting a density model [126] and learning a dynamics model [127].”
Influence-Based Multi-Agent Exploration
Wang, T., Wang, J., Wu, Y., & Zhang, C. (2019) · arXiv · doi:10.48550/arxiv.1910.05512 · Cited by 69
Successor Uncertainties: Exploration and Uncertainty in Temporal Difference Learning
Janz, D., Hron, J., Mazur, P., Hofmann, K., Hernández-Lobato, J. M., & Tschiatschek, S. (2018) · arXiv · doi:10.48550/arxiv.1810.06530 · Cited by 23
Cited by
Ajagekar, A., Decardi-Nelson, B., & You, F. (2024) · Applied Energy · doi:10.1016/j.apenergy.2023.122349 · Cited by 82
Kadi, H. A., & Terzić, K. (2023) · Sensors · doi:10.3390/s23052389 · Cited by 27
Metadata
- Authors
- Hao, J., Yang, T., Tang, H., Bai, C., Liu, J., Meng, Z., Liu, P., & Wang, Z.
- Venue
- IEEE Transactions on Neural Networks and Learning Systems
- Published
- 2024-07
- Volume
- 35
- Issue
- 7
- Pages
- 8762-8782
- ISSN
- 2162-237X (ISSN-L)
- 2162-2388
- OpenAlex
- W4319165238
- Cited by
- 189
- References
- 44
- FWCI
- 16.43
- Type
- journal article
- Language
- en
- Open access
- green
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.
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.