Multi-Task Reinforcement Learning with Language-Encoded Gated Policy Networks
This addresses the problem of efficiently learning and reusing skills across diverse tasks in reinforcement learning, though it is incremental as it builds on existing multi-task RL methods with language conditioning.
The paper tackled multi-task reinforcement learning by introducing Lexical Policy Networks (LEXPOL), a language-conditioned architecture that encodes task metadata to select or blend sub-policies, achieving matching or superior success rates and sample efficiency on MetaWorld benchmarks without task-specific retraining.
Multi-task reinforcement learning often relies on task metadata -- such as brief natural-language descriptions -- to guide behavior across diverse objectives. We present Lexical Policy Networks (LEXPOL), a language-conditioned mixture-of-policies architecture for multi-task RL. LEXPOL encodes task metadata with a text encoder and uses a learned gating module to select or blend among multiple sub-policies, enabling end-to-end training across tasks. On MetaWorld benchmarks, LEXPOL matches or exceeds strong multi-task baselines in success rate and sample efficiency, without task-specific retraining. To analyze the mechanism, we further study settings with fixed expert policies obtained independently of the gate and show that the learned language gate composes these experts to produce behaviors appropriate to novel task descriptions and unseen task combinations. These results indicate that natural-language metadata can effectively index and recombine reusable skills within a single policy.