Learning Context-aware Task Reasoning for Efficient Meta-reinforcement Learning

TL;DR

Meta-reinforcement learning often faces sampling inefficiency and meta-overfitting, especially with off-policy methods. The paper presents CASTER, a three-task decomposition into task-exploration, task-inference, and task-fulfillment, featuring dual agents and a context-aware graph encoder, trained with a variational EM objective to infer a latent task variable and customize task-specific policies. It introduces a self-supervised reward shaping mechanism for exploration and a Latent Graph Neural Network–based encoder to model dependencies among experiences, enabling efficient task inference. Across diverse benchmarks, CASTER demonstrates improved test-time sample efficiency, higher asymptotic performance, and reduced train-test distribution mismatch, indicating strong generalization under varied task distributions.

Abstract

Despite recent success of deep network-based Reinforcement Learning (RL), it remains elusive to achieve human-level efficiency in learning novel tasks. While previous efforts attempt to address this challenge using meta-learning strategies, they typically suffer from sampling inefficiency with on-policy RL algorithms or meta-overfitting with off-policy learning. In this work, we propose a novel meta-RL strategy to address those limitations. In particular, we decompose the meta-RL problem into three sub-tasks, task-exploration, task-inference and task-fulfillment, instantiated with two deep network agents and a task encoder. During meta-training, our method learns a task-conditioned actor network for task-fulfillment, an explorer network with a self-supervised reward shaping that encourages task-informative experiences in task-exploration, and a context-aware graph-based task encoder for task inference. We validate our approach with extensive experiments on several public benchmarks and the results show that our algorithm effectively performs exploration for task inference, improves sample efficiency during both training and testing, and mitigates the meta-overfitting problem.

Figures (9)

• Figure 1: Context-aware task reasoning for RL adaptation. We separate the task into task-exploration, task inference and task-fulfillment. The explorer interacts with the environment to collect experiences for the task encoder to update the belif of task. After an iterative process of $K-1$ rounds, the task encoder takes all the collections and gives the final task hypothesis to adapt the actor.
• Figure 2: The task encoder.The first aggregation constructs a bipartite graph with full connections from the $n$ nodes in $V^i$ to the $c$ latent nodes in $V^h$. Self-attention operates on $V^h$, which are assembled to one latent node in the second aggregation.
• Figure 3: Training efficiency. The test-task performance versus the number of interactions with the environment during meta-training. The dash lines represent the asymptotic performance of each method.
• Figure 4: Testing efficiency. The x-axis denotes the number of trajectories used as adaptation data.
• Figure 5: Overfitting in off-policy meta-RL. Each column in the plot corresponds to a different environment. We pick three environments most prone to meta-overfitting, i.e., 'cheetah-vel', 'ant-goal' and 'point-robot' from left to right.