Goal2Skill: Long-Horizon Manipulation with Adaptive Planning and Reflection

TL;DR

Goal2Skill introduces a dual-system framework for long-horizon embodied manipulation, combining high-level planning with low-level visuomotor control to improve robustness and success rates in complex tasks.

Contribution

The paper presents a novel dual-system approach that separates semantic planning from motor execution, enabling memory-aware reasoning and adaptive recovery in long-horizon tasks.

Findings

Achieved a 32.4% success rate on RMBench tasks, outperforming the 9.8% of the best baseline.

Structured memory and closed-loop recovery significantly improve task success.

The framework effectively handles partial observability, occlusions, and multi-stage dependencies.

Abstract

Recent vision-language-action (VLA) systems have demonstrated strong capabilities in embodied manipulation. However, most existing VLA policies rely on limited observation windows and end-to-end action prediction, which makes them brittle in long-horizon, memory-dependent tasks with partial observability, occlusions, and multi-stage dependencies. Such tasks require not only precise visuomotor control, but also persistent memory, adaptive task decomposition, and explicit recovery from execution failures. To address these limitations, we propose a dual-system framework for long-horizon embodied manipulation. Our framework explicitly separates high-level semantic reasoning from low-level motor execution. A high-level planner, implemented as a VLM-based agentic module, maintains structured task memory and performs goal decomposition, outcome verification, and error-driven correction. A low-level executor, instantiated as a VLA-based visuomotor controller, carries out each sub-task through diffusion-based action generation conditioned on geometry-preserving filtered observations. Together, the two systems form a closed loop between planning and execution, enabling memory-aware reasoning, adaptive replanning, and robust online recovery. Experiments on representative RMBench tasks show that the proposed framework substantially outperforms representative baselines, achieving a 32.4% average success rate compared with 9.8% for the strongest baseline. Ablation studies further confirm the importance of structured memory and closed-loop recovery for long-horizon manipulation.