skip to main content

Attention:

The NSF Public Access Repository (PAR) system and access will be unavailable from 11:00 PM ET on Thursday, February 13 until 2:00 AM ET on Friday, February 14 due to maintenance. We apologize for the inconvenience.


Title: Constrained Variational Policy Optimization for Safe Reinforcement Learning
afe reinforcement learning (RL) aims to learn policies that satisfy certain constraints before deploying them to safety-critical applications. Previous primal-dual style approaches suffer from instability issues and lack optimality guarantees. This paper overcomes the issues from the perspective of probabilistic inference. We introduce a novel Expectation-Maximization approach to naturally incorporate constraints during the policy learning: 1) a provable optimal non-parametric variational distribution could be computed in closed form after a convex optimization (E-step); 2) the policy parameter is improved within the trust region based on the optimal variational distribution (M-step). The proposed algorithm decomposes the safe RL problem into a convex optimization phase and a supervised learning phase, which yields a more stable training performance. A wide range of experiments on continuous robotic tasks shows that the proposed method achieves significantly better constraint satisfaction performance and better sample efficiency than baselines. The code is available at https://github.com/liuzuxin/cvpo-safe-rl.  more » « less
Award ID(s):
2047454
PAR ID:
10416733
Author(s) / Creator(s):
; ; ; ; ; ;
Date Published:
Journal Name:
Proceedings of the 39th International Conference on Machine Learning
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Existing offline reinforcement learning (RL) methods face a few major challenges, particularly the distributional shift between the learned policy and the behavior policy. Offline Meta-RL is emerging as a promising approach to address these challenges, aiming to learn an informative meta-policy from a collection of tasks. Nevertheless, as shown in our empirical studies, offline Meta-RL could be outperformed by offline single-task RL methods on tasks with good quality of datasets, indicating that a right balance has to be delicately calibrated between "exploring" the out-of-distribution state-actions by following the meta-policy and "exploiting" the offline dataset by staying close to the behavior policy. Motivated by such empirical analysis, we propose model-based offline ta-RL with regularized policy optimization (MerPO), which learns a meta-model for efficient task structure inference and an informative meta-policy for safe exploration of out-of-distribution state-actions. In particular, we devise a new meta-Regularized model-based Actor-Critic (RAC) method for within-task policy optimization, as a key building block of MerPO, using both conservative policy evaluation and regularized policy improvement; and the intrinsic tradeoff therein is achieved via striking the right balance between two regularizers, one based on the behavior policy and the other on the meta-policy. We theoretically show that the learnt policy offers guaranteed improvement over both the behavior policy and the meta-policy, thus ensuring the performance improvement on new tasks via offline Meta-RL. Our experiments corroborate the superior performance of MerPO over existing offline Meta-RL methods. 
    more » « less
  2. The actor-critic RL is widely used in various robotic control tasks. By viewing the actor-critic RL from the perspective of variational inference (VI), the policy network is trained to obtain the approximate posterior of actions given the optimality criteria. However, in practice, the actor-critic RL may yield suboptimal policy estimates due to the amortization gap and insufficient exploration. In this work, inspired by the previous use of Hamiltonian Monte Carlo (HMC) in VI, we propose to integrate the policy network of actor-critic RL with HMC, which is termed as Hamiltonian Policy. As such we propose to evolve actions from the base policy according to HMC, and our proposed method has many benefits. First, HMC can improve the policy distribution to better approximate the posterior and hence reduce the amortization gap. Second, HMC can also guide the exploration more to the regions of action spaces with higher Q values, enhancing the exploration efficiency. Further, instead of directly applying HMC into RL, we propose a new leapfrog operator to simulate the Hamiltonian dynamics. Finally, in safe RL problems, we find that the proposed method can not only improve the achieved return, but also reduce safety constraint violations by discarding potentially unsafe actions. With comprehensive empirical experiments on continuous control baselines, including MuJoCo and PyBullet Roboschool, we show that the proposed approach is a data-efficient and easy-to-implement improvement over previous actor-critic methods. 
    more » « less
  3. Reinforcement learning (RL) has recently shown promise in solving difficult numerical problems and has discovered non-intuitive solutions to existing problems. This study investigates the ability of a general RL agent to find an optimal control strategy for spacecraft attitude control problems. Two main types of Attitude Control Systems (ACS) are presented. First, the general ACS problem with full actuation is considered, but with saturation constraints on the applied torques, representing thruster-based ACSs. Second, an attitude control problem with reaction wheel based ACS is considered, which has more constraints on control authority. The agent is trained using the Proximal Policy Optimization (PPO) RL method to obtain an attitude control policy. To ensure robustness, the inertia of the satellite is unknown to the control agent and is randomized for each simulation. To achieve efficient learning, the agent is trained using curriculum learning. We compare the RL based controller to a QRF (quaternion rate feedback) attitude controller, a well-established state feedback control strategy. We investigate the nominal performance and robustness with respect to uncertainty in system dynamics. Our RL based attitude control agent adapts to any spacecraft mass without needing to re-train. In the range of 0.1 to 100,000 kg, our agent achieves 2% better performance to a QRF controller tuned for the same mass range, and similar performance to the QRF controller tuned specifically for a given mass. The performance of the trained RL agent for the reaction wheel based ACS achieved 10 higher better reward then that of a tuned QRF controller 
    more » « less
  4. Safe reinforcement learning (RL) has been recently employed to train a control policy that maximizes the task reward while satisfying safety constraints in a simulated secure cyber-physical environment. However, the vulnerability of safe RL has been barely studied in an adversarial setting. We argue that understanding the safety vulnerability of learned control policies is essential to achieve true safety in the physical world. To fill this research gap, we first formally define the adversarial safe RL problem and show that the optimal policies are vulnerable under observation perturbations. Then, we propose novel safety violation attacks that induce unsafe behaviors by adversarial models trained using reversed safety constraints. Finally, both theoretically and experimentally, we show that our method is more effective in violating safety than existing adversarial RL works which just seek to decrease the task reward, instead of violating safety constraints. 
    more » « less
  5. Reinforcement learning (RL) in low-data and risk-sensitive domains requires performant and flexible deployment policies that can readily incorporate constraints during deployment. One such class of policies are the semi-parametric H-step lookahead policies, which select actions using trajectory optimization over a dynamics model for a fixed horizon with a terminal value function. In this work, we investigate a novel instantiation of H-step lookahead with a learned model and a terminal value function learned by a model-free off-policy algorithm, named Learning Off-Policy with Online Planning (LOOP). We provide a theoretical analysis of this method, suggesting a tradeoff between model errors and value function errors and empirically demonstrate this tradeoff to be beneficial in deep reinforcement learning. Furthermore, we identify the "Actor Divergence" issue in this framework and propose Actor Regularized Control (ARC), a modified trajectory optimization procedure. We evaluate our method on a set of robotic tasks for Offline and Online RL and demonstrate improved performance. We also show the flexibility of LOOP to incorporate safety constraints during deployment with a set of navigation environments. We demonstrate that LOOP is a desirable framework for robotics applications based on its strong performance in various important RL settings. 
    more » « less