A Simple Mixture Policy Parameterization for Improving Sample Efficiency of CVaR Optimization

Reinforcement learning algorithms utilizing policy gradients (PG) to optimize Conditional Value at Risk (CVaR) face significant challenges with sample inefficiency, hindering their practical applications. This inefficiency stems from two main facts: a focus on tail-end performance that overlooks many sampled trajectories, and the potential of gradient vanishing when the lower tail of the return distribution is overly flat. To address these challenges, we propose a simple mixture policy parameterization. This method integrates a risk-neutral policy with an adjustable policy to form a risk-averse policy. By employing this strategy, all collected trajectories can be utilized for policy updating, and the issue of vanishing gradients is counteracted by stimulating higher returns through the risk-neutral component, thus lifting the tail and preventing flatness. Our empirical study reveals that this mixture parameterization is uniquely effective across a variety of benchmark domains. Specifically, it excels in identifying risk-averse CVaR policies in some Mujoco environments where the traditional CVaR-PG fails to learn a reasonable policy.

An Alternative to Variance: Gini Deviation for Risk-averse Policy Gradient

Restricting the variance of a policy's return is a popular choice in risk-averse Reinforcement Learning (RL) due to its clear mathematical definition and easy interpretability. Traditional methods directly restrict the total return variance. Recent methods restrict the per-step reward variance as a proxy. We thoroughly examine the limitations of these variance-based methods, such as sensitivity to numerical scale and hindering of policy learning, and propose to use an alternative risk measure, Gini deviation, as a substitute. We study various properties of this new risk measure and derive a policy gradient algorithm to minimize it. Empirical evaluation in domains where risk-aversion can be clearly defined, shows that our algorithm can mitigate the limitations of variance-based risk measures and achieves high return with low risk in terms of variance and Gini deviation when others fail to learn a reasonable policy.

Benchmarking Constraint Inference in Inverse Reinforcement Learning

When deploying Reinforcement Learning (RL) agents into a physical system, we must ensure that these agents are well aware of the underlying constraints. In many real-world problems, however, the constraints followed by expert agents (e.g., humans) are often hard to specify mathematically and unknown to the RL agents. To tackle these issues, Constraint Inverse Reinforcement Learning (CIRL) considers the formalism of Constrained Markov Decision Processes (CMDPs) and estimates constraints from expert demonstrations by learning a constraint function. As an emerging research topic, CIRL does not have common benchmarks, and previous works tested their algorithms with hand-crafted environments (e.g., grid worlds). In this paper, we construct a CIRL benchmark in the context of two major application domains: robot control and autonomous driving. We design relevant constraints for each environment and empirically study the ability of different algorithms to recover those constraints based on expert trajectories that respect those constraints. To handle stochastic dynamics, we propose a variational approach that infers constraint distributions, and we demonstrate its performance by comparing it with other CIRL baselines on our benchmark. The benchmark, including the information for reproducing the performance of CIRL algorithms, is publicly available at https://github.com/Guiliang/CIRL-benchmarks-public

Learning Selective Communication for Multi-Agent Path Finding

Learning communication via deep reinforcement learning (RL) or imitation learning (IL) has recently been shown to be an effective way to solve Multi-Agent Path Finding (MAPF). However, existing communication based MAPF solvers focus on broadcast communication, where an agent broadcasts its message to all other or predefined agents. It is not only impractical but also leads to redundant information that could even impair the multi-agent cooperation. A succinct communication scheme should learn which information is relevant and influential to each agent's decision making process. To address this problem, we consider a request-reply scenario and propose Decision Causal Communication (DCC), a simple yet efficient model to enable agents to select neighbors to conduct communication during both training and execution. Specifically, a neighbor is determined as relevant and influential only when the presence of this neighbor causes the decision adjustment on the central agent. This judgment is learned only based on agent's local observation and thus suitable for decentralized execution to handle large scale problems. Empirical evaluation in obstacle-rich environment indicates the high success rate with low communication overhead of our method.

Distributed Heuristic Multi-Agent Path Finding with Communication

Multi-Agent Path Finding (MAPF) is essential to large-scale robotic systems. Recent methods have applied reinforcement learning (RL) to learn decentralized polices in partially observable environments. A fundamental challenge of obtaining collision-free policy is that agents need to learn cooperation to handle congested situations. This paper combines communication with deep Q-learning to provide a novel learning based method for MAPF, where agents achieve cooperation via graph convolution. To guide RL algorithm on long-horizon goal-oriented tasks, we embed the potential choices of shortest paths from single source as heuristic guidance instead of using a specific path as in most existing works. Our method treats each agent independently and trains the model from a single agent's perspective. The final trained policy is applied to each agent for decentralized execution. The whole system is distributed during training and is trained under a curriculum learning strategy. Empirical evaluation in obstacle-rich environment indicates the high success rate with low average step of our method.