VoLo: A Physical Orchestrator for Open-Vocabulary Long-Horizon Manipulation

2026-06-05

AI Summary

This work addresses the challenges of flexible instruction understanding, complex scene reasoning, and real-time fault-tolerant execution in open-vocabulary, long-horizon robotic manipulation by proposing VoLoAgent, a closed-loop embodied agent. VoLoAgent introduces a novel “physical orchestration” paradigm that synergistically integrates vision-language models (VLMs) and vision-language-action models (VLAs/WAMs) as interruptible and intervenable tools, dynamically invoking action primitives for planning, monitoring, and failure recovery. Evaluated on the newly introduced RoboVoLo benchmark, VoLoAgent substantially outperforms systems based solely on VLAs or VLMs as well as conventional tool-augmented approaches. Its efficacy and robustness are further validated through high-fidelity simulation and real-world robotic experiments.

Abstract

Open-vocabulary long-horizon manipulation requires robots to reason over flexible instructions and complex multi-object scenes while adaptively planning, executing, monitoring, and recovering from failures. We address these demands with a closed agent loop in which a VLM orchestrates heterogeneous robot capabilities as interruptible tools. Unlike in virtual AI agents, the timing of decisions, actions and tool calls is important in a physical world that does not pause for reasoning. We refer to this setting as Physical Orchestration, and propose VoLoAgent, a VLM that plans, monitors, and recovers by treating a VLA/WAM as an interruptible tool it steers mid-rollout alongside vision models and action primitives. To evaluate these long-horizon capabilities, we introduce RoboVoLo, a high-fidelity benchmark for open-vocabulary long-horizon manipulation across common sense, memory/state tracking, complex references, and world knowledge, with both task-level success and failure-mode diagnostics. Experiments show VoLoAgent substantially outperforms single VLA/VLM or tool-based systems, with validation on real-robot experiments. Project page: https://chicychen.github.io/VoLo/

Problem

• open-vocabulary manipulation
• long-horizon manipulation
• physical orchestration
• robot failure recovery
• complex multi-object scenes

Innovation

• Physical Orchestration
• Open-Vocabulary Manipulation
• Interruptible Tools
• Long-Horizon Planning
• Vision-Language Models