Timezone: »

 
Plan, Eliminate, and Track --- Language Models are Good Teachers for Embodied Agents.
Yue Wu · So Yeon Min · Yonatan Bisk · Ruslan Salakhutdinov · Amos Azaria · Yuanzhi Li · Tom Mitchell · Shrimai Prabhumoye

@
in Knowledge and Logical Reasoning in the Era of Data-driven Learning »

Pre-trained large language models (LLMs) capture procedural knowledge about the world. Recent work has leveraged LLM's ability to generate abstract plans to simplify challenging control tasks, either by action scoring, or action modeling (fine-tuning). However, the transformer architecture inherits several constraints that make it difficult for the LLM to directly serve as the agent: e.g. limited input lengths, fine-tuning inefficiency, bias from pre-training, and incompatibility with non-text environments. To maintain compatibility with a low-level trainable actor, we propose to instead use the knowledge in LLMs to simplify the control problem, rather than solving it. We propose the Plan, Eliminate, and Track (PET) framework. The Plan module translates a task description into a list of high-level sub-tasks. The Eliminate module masks out irrelevant objects and receptacles from the observation for the current sub-task. Finally, the Track module determines whether the agent has accomplished each sub-task. On the AlfWorld instruction following benchmark, the PET framework leads to a significant 15% improvement over SOTA for generalization to human goal specifications.

Author Information

Yue Wu (Carnegie Mellon University)
So Yeon Min (Carngie Mellon University)
Yonatan Bisk (Carnegie Mellon University)
Ruslan Salakhutdinov (Carnegie Mellen University)
Amos Azaria (Ariel University)
Yuanzhi Li (CMU)
Tom Mitchell (CMU)
Shrimai Prabhumoye (NVIDIA)

More from the Same Authors