Natural Actor-Critic for Robust Reinforcement Learning with Function Approximation

We study robust reinforcement learning (RL) with the goal of determining a well-performing policy that is robust against model mismatch between the training simulator and the testing environment. Previous policy-based robust RL algorithms mainly focus on the tabular setting under uncertainty sets that facilitate robust policy evaluation, but are no longer tractable when the number of states scales up. To this end, we propose two novel uncertainty set formulations, one based on double sampling and the other on an integral probability metric. Both make large-scale robust RL tractable even when one only has access to a simulator. We propose a robust natural actor-critic (RNAC) approach that incorporates the new uncertainty sets and employs function approximation. We provide finite-time convergence guarantees for the proposed RNAC algorithm to the optimal robust policy within the function approximation error. Finally, we demonstrate the robust performance of the policy learned by our proposed RNAC approach in multiple MuJoCo environments and a real-world TurtleBot navigation task.

Paper

References (70)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer 4Et95/10 · confidence 4/52023-07-04

Summary

This paper studies the actor-critic approach for robust RL. Especially, a Double-Sampling Uncertainty Set and an Integral Probability Metric Uncertainty Set are developed to overcome the curse of problem scale. A robust natural Actor-Critic algorithm is then proposed with convergence results. A significant number of experiments are designed.

Strengths

The paper is well-written and clear in general. The design of the two uncertainty sets is new and novel, which shows advantages under large-scale problems.

Weaknesses

1. The motivation for designing the two uncertainty sets is somehow unclear to me. I understand there is uncertainty in the sets designed, but don't understand the motivation of this uncertainty set. In lines 640-646, the authors explain the uncertainty set contains transition kernels that are perturbed from the uniform distribution, this explanation seems unclear to me, and I can't understand the motivation for such a definition. 2. One of the critical problems in studying robust RL with function approximation is the contraction of the approximated Bellman operator. The approach used in this paper is similar to the previous ones, i.e., use conditions on the radius of the uncertainty set and discount factor. This hence reduces the novelty and contribution of the paper.

Questions

Can you explain the motivation for the designing of the uncertainty sets? In what sense do they imply robustness and uncertainty?

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

See the parts above.

Reviewer r1LZ7/10 · confidence 3/52023-07-05

Summary

This paper tackles robust reinforcement learning in large state spaces, where the transition kernel is accessible only in a nominal setting. The authors demonstrate that the $f$-divergence, R-contamination, and the $l_{p}$ norm are computationally infeasible in the context of robust RL for large state spaces. To overcome this limitation, the authors propose two new tractable uncertainty set formulations suitable for large dimensions: double sampling (DS) and the integral probability metric (IPM). DS involves independently and identically distributed state sampling following a transition kernel $p^o_{s,a}$. The IPM corresponds to the robust Bellman operator, but with a regularization term on the norm of weights. Both approaches are enabling Robust RL in scenarios previously hindered by computational complexity. The paper introduces a new algorithm, the Robust Natural Actor Critic (RNAC), for training both a critic and an actor for the proposed IPM and double sampling uncertainty sets. The authors provide convergence guarantees for the RNAC algorithm. Finally, the authors demonstrate the efficency of their approach via two applications: one involving the suite of MuJoCo environments and the other, a real-world robotics application. These practical applications lend credence to the theoretical contributions and the robustness of the proposed approach.

Strengths

- The authors provide two straightforward and computationally feasible uncertainty set formulations for large state spaces. - The paper offers substantial theoretical contributions. - The real-world robotics application lends credibility to the paper's robustness claims.

Weaknesses

- The convergence guarantees are valid for $(s,a)$-rectangular uncertainty sets, which is rather limiting. However, it is commendable that the authors have acknowledged this limitation in their work and suggested it as an avenue for future research. - The paper seems incomplete without the appendix. The need to constantly refer to the appendix disrupts the flow of reading.

Questions

- In lines 175-177, why did you not apply regularization to the bias parameter in the last layer? Moreover, this seems inconsistent with the provided implementation (lines 201-206 in the `RNAC-ppo/ppo_continious.py` file), where regularization is applied to all the critic's layers. - Doesn't the IPM's proposed regularization ultimately reduce the Lipschitz constant of the learned function, making the agent less sensitive to state variations?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

4 excellent

Presentation

2 fair

Contribution

3 good

Limitations

The authors mention their limitations as future research directions. We appreciate the authors for being upfront.

Reviewer haxo6/10 · confidence 2/52023-07-07

Summary

This paper studies the sim-to-real transfer problem. It extends the learning of a robust policy by using the framework of robust Markov decision processes (RMDPs). It extends this paradigm to large state and action spaces using two uncertainty set formulations: double sampling, and integral probability metric. These formulations are then used in the proposed algorithm robust natural actor critic (RNAC). RNAC is tested in MuJoCo as well as on a real robot.

Strengths

* Proposed uncertainty sets as well as the RNAC algorithm seem like practical steps forward in sim-to-real transfer with robustness approaches. * The theoretical analysis seems interesting

Weaknesses

* The experiments compare only to PPO. They should compare to other sim-to-real methods such as dynamics randomization [1] or action noise envelope [2]. * The related works being relegated to the appendix seems like a red flag. The authors should better organize the paper to include related works and comparisons to the main paper. ### References [1] Peng, X.B., Andrychowicz, M., Zaremba, W. and Abbeel, P., 2018, May. Sim-to-real transfer of robotic control with dynamics randomization. In 2018 IEEE international conference on robotics and automation (ICRA) (pp. 3803-3810). IEEE. [2] Jakobi, N., Husbands, P. and Harvey, I., 1995. Noise and the reality gap: The use of simulation in evolutionary robotics. In Advances in Artificial Life: Third European Conference on Artificial Life Granada, Spain, June 4–6, 1995 Proceedings 3 (pp. 704-720). Springer Berlin Heidelberg. ================================ The additional experiments alleviate my concerns about the experimental analysis.

Questions

* In the second to last paragraph of Section 2, when describing the single worst-case kernel, is this kernel constant for the entire state space or varying? * In the double sampling setup, does $\sum \alpha = 1$ in Equation (3)?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

2: You are willing to defend your assessment, but it is quite likely that you did not understand the central parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

n/a

Reviewer Fq666/10 · confidence 1/52023-08-01

Summary

RL methods trained on simulators suffered from generalization problems because of the "simulation-to-reality-gap". Previous works proposed robust RL methods in a tabular setting, with limited search spaces. The paper aims to develop a computationally tractable robust RL algorithm with large search spaces. To this end, the paper proposed two novel uncertainty sets and the first policy-based approach for robust RL with provable convergence guarantees.

Strengths

The paper studies a critical problem. Several technical contributions are proposed to devise a robust policy-based RL method with a large search space. The theoretical analysis seems solid.

Weaknesses

The paper aims to devise a robust RL method. More real-world experiments are expected to demonstrate the robustness of the proposed methods. --------------------- After reading the rebuttal, my main concerns were addressed.

Questions

- Can previous tabulated-based robust RL methods be applied to the experiments in Sec. 7? If so, what about the comparison results? On the other hand, can the author provide some experiments to directly compare the proposed method with large search space to the previous one with limited search space to demonstrate the benefits of the proposed method?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

1: Your assessment is an educated guess. The submission is not in your area or the submission was difficult to understand. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

The paper has discussed its limitations.

Reviewer r1LZ2023-08-13

Response to rebuttal

The rebuttal addressed my concerns. I'll keep my rating.

Authorsrebuttal2023-08-16

We are pleased that our rebuttal has addressed the reviewer's concerns. We greatly appreciate the reviewer's recognition of our work.

Reviewer Fq662023-08-14

The rebuttal has addressed my main concerns. I'll keep my rating.

Authorsrebuttal2023-08-17

We thank the reviewer for the valuable comments on experimental evaluation and we are pleased to know that our rebuttal addresses the reviewer's concerns. Please let us know if you have any further questions. We will be happy to answer them. If you find our response satisfying, we wonder if you could kindly consider raising the score rating of our work?

Reviewer haxo2023-08-15

Thank you for the response

I thank the authors for their response and for the additional experiments taking into account the suggestions put forth in the review. These experiments now certainly showcase the effectiveness of the proposed technique better. I also thank the authors for answering the questions in the review. The related works could be better presented, but the authors' possible modifications might make the comparison sufficient. With these points, I will be revising my score.

Authorsrebuttal2023-08-17

We are pleased that our rebuttal addresses the reviewer's concerns. We thank the reviewer for the valuable suggestions and recognition of our work. We would be glad to take action on any further suggestions from the reviewer to make our presentation better.

Reviewer 4Et92023-08-21

The response solves my concerns, and I hence increase my score.

Authorsrebuttal2023-08-21

We thank the reviewer for carefully reading our rebuttal and raising the score. We will add the above clarification in our final version.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC