A Few Words Go a Long Way: Language Guided Robot Policy Synthesis

2026-07-26

AI Summary

This work addresses the limited interpretability and correctability of existing vision–language–action (VLA) models, which often act as black-box policies and suffer from cascading failures under distribution shifts. To overcome these limitations, the authors propose the ARCHITECT framework, which formulates robotic policy generation as an interactive program synthesis task. Leveraging a large language model, ARCHITECT synthesizes modular programs and iteratively refines them at the code level through natural language feedback. The approach integrates execution-trace-anchored feedback, constructs a persistent skill library for long-term learning and behavior reuse, and unifies perception and control tools within a coherent architecture. Evaluated on the Franka Panda platform, the system substantially outperforms current VLA and program synthesis baselines, successfully completing complex long-horizon tasks and enabling skill transfer with progressively reduced human intervention as tasks evolve.

Abstract

While vision-language-action models have demonstrated impressive zero-shot manipulation capabilities, they remain fundamentally black box policies that are difficult to interpret, adapt, or correct when they inevitably fail. In this work, we propose ARCHITECT, a framework that treats robot policy acquisition as an interactive program synthesis task. ARCHITECT leverages the reasoning capabilities of LLM coding agents to synthesize modular robot programs that utilize a suite of perception and control tools. Unlike end-to-end models where distribution shift leads to unpredictable, cascading failures, our modular architecture allows users to isolate failures and localize feedback at the level of abstraction required. We introduce an iterative process where a human supervisor provides natural language corrections to steer the policy. These corrections are grounded in the policy code by program execution traces and distilled into a persistent skill library, a form of long-term in-context learning which enables the agent to accumulate a repertoire of reusable, interpretable behaviors. In a benchmark evaluation on a Franka Panda robot, ARCHITECT outperforms state-of-the-art VLA models and program synthesis baselines on complex, long-horizon tasks, including articulated object manipulation and cloth folding. Our results demonstrate that the synthesized skill library enables the system to transfer to novel tasks with decreasing human intervention, providing a steerable and data-efficient alternative to black-box robot learning. Website: https://robo-architect.github.io/

Problem

• vision-language-action models
• black-box policies
• policy interpretability
• failure correction
• distribution shift

Innovation

• program synthesis
• modular robot policy
• language-guided learning
• skill library
• interpretable robotics