Bimanual Robot Manipulation via Multi-Agent In-Context Learning

TL;DR

This paper introduces BiCICLe, a novel framework enabling large language models to perform few-shot bimanual robot manipulation without fine-tuning, by framing the task as a multi-agent problem with iterative refinement.

Contribution

BiCICLe is the first approach to enable standard LLMs to handle bimanual manipulation tasks through multi-agent in-context learning and iterative trajectory refinement.

Findings

Achieves up to 71.1% success rate on TWIN benchmark tasks.

Outperforms the best training-free baseline by 6.7 percentage points.

Demonstrates strong few-shot generalization to new tasks.

Abstract

Language Models (LLMs) have emerged as powerful reasoning engines for embodied control. In particular, In-Context Learning (ICL) enables off-the-shelf, text-only LLMs to predict robot actions without any task-specific training while preserving their generalization capabilities. Applying ICL to bimanual manipulation remains challenging, as the high-dimensional joint action space and tight inter-arm coordination constraints rapidly overwhelm standard context windows. To address this, we introduce BiCICLe (Bimanual Coordinated In-Context Learning), the first framework that enables standard LLMs to perform few-shot bimanual manipulation without fine-tuning. BiCICLe frames bimanual control as a multi-agent leader-follower problem, decoupling the action space into sequential, conditioned single-arm predictions. This naturally extends to Arms' Debate, an iterative refinement process, and to the introduction of a third LLM-as-Judge to evaluate and select the most plausible coordinated trajectories. Evaluated on 13 tasks from the TWIN benchmark, BiCICLe achieves up to 71.1% average success rate, outperforming the best training-free baseline by 6.7 percentage points and surpassing most supervised methods. We further demonstrate strong few-shot generalization on novel tasks.