Abstract
In environments with continuous state and action spaces, state-of-the-art actor-critic reinforcement learning algorithms can solve very complex problems, yet can also fail in environments that seem trivial, but the reason for such failures is still poorly understood. In this paper, we contribute a formal explanation of these failures in the particular case of sparse reward and deterministic environments. First, using a very elementary control problem, we illustrate that the learning process can get stuck into a fixed point corresponding to a poor solution, especially when the reward is not found very early. Then, generalizing from the studied example, we provide a detailed analysis of the underlying mechanisms which results in a new understanding of one of the convergence regimes of these algorithms.
This work was partially supported by the French National Research Agency (ANR), Project ANR-18-CE33-0005 HUSKI.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Similar content being viewed by others
Notes
- 1.
10% of steps are governed by probabilistic noise, of which at least 2% are the first episode step, of which 50% are steps going to the left and leading to the reward.
- 2.
References
Achiam, J., Knight, E., Abbeel, P.: Towards characterizing divergence in deep q-learning. arXiv:1903.08894 (2019)
Ahmed, Z., Roux, N.L., Norouzi, M., Schuurmans, D.: Understanding the impact of entropy on policy optimization. arXiv:1811.11214 (2019)
Baird, L.C., Klopf, A.H.: Technical Report WL-TR-93-1147. Wright-Patterson AIr Force Base, Ohio, Wright Laboratory (1993)
Boyan, J.A., Moore, A.W.: Generalization in reinforcement learning: safely approximating the value function. In: Advances in Neural Information Processing Systems, pp. 369–376 (1995)
Colas, C., Sigaud, O., Oudeyer, P.Y.: GEP-PG: Decoupling Exploration and Exploitation in Deep Reinforcement Learning Algorithms. arXiv:1802.05054 (2018)
Fortunato, M., et al.: Noisy Networks for Exploration. arXiv:1706.10295 (2017)
Fujimoto, S., van Hoof, H., Meger, D.: Addressing Function Approximation Error in Actor-Critic Methods. ICML (2018)
Fujimoto, S., Meger, D., Precup, D.: Off-Policy Deep Reinforcement Learning without Exploration. arXiv:1812.02900 (2018)
Geist, M., Pietquin, O.: Parametric value function approximation: a unified view. In: ADPRL 2011, Paris, France, pp. 9–16 (2011)
van Hasselt, H., Doron, Y., Strub, F., Hessel, M., Sonnerat, N., Modayil, J.: Deep Reinforcement Learning and the Deadly Triad. arXiv:1812.02648 (2018)
Lillicrap, T.P., et al.: Continuous control with deep reinforcement learning. arXiv:1509.02971 (2015)
Mnih, V., et al.: Playing Atari with Deep Reinforcement Learning. arXiv:1312.5602 (2013)
Mnih, V., et al.: Human-level control through deep reinforcement learning. Nature 518(7540), 529–533 (2015)
Plappert, M., et al.: Parameter space noise for exploration. arXiv preprint arXiv:1706.01905 (2017)
Silver, D., Lever, G., Heess, N., Degris, T., Wierstra, D., Riedmiller, M.: Deterministic policy gradient algorithms. In: International Conference on Machine Learning, pp. 387–395 (2014)
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press, Cambridge (2018)
Tsitsiklis, J.N., Van Roy, B.: Analysis of temporal-difference learning with function approximation. In: Advances in Neural Information Processing Systems, pp. 1075–1081 (1997)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Matheron, G., Perrin, N., Sigaud, O. (2020). Understanding Failures of Deterministic Actor-Critic with Continuous Action Spaces and Sparse Rewards. In: Farkaš, I., Masulli, P., Wermter, S. (eds) Artificial Neural Networks and Machine Learning – ICANN 2020. ICANN 2020. Lecture Notes in Computer Science(), vol 12397. Springer, Cham. https://doi.org/10.1007/978-3-030-61616-8_25
Download citation
DOI: https://doi.org/10.1007/978-3-030-61616-8_25
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-61615-1
Online ISBN: 978-3-030-61616-8
eBook Packages: Computer ScienceComputer Science (R0)