cs.RO提交于 2026-10-30已译

ELEVATE：从任务失败中获取四足操作智能体控制器

ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation

Shiqi Liu · Radu Corcodel · Ding Zhao

中文摘要

四足操作机器人可通过全身交互扩展工作空间，但任务规划智能体的能力受限于其可用控制器。ELEVATE 框架利用任务程序精炼过程中的持续失败来识别并获取缺失的全身控制能力。当失败表明缺少某种物理能力时，编码智能体构建并迭代精炼一个强化学习问题，指定行为特定奖励、课程和终止条件。验证通过的控制器加入能力库，与现有控制器组合后重试任务。从单个预训练移动-操作控制器出发，ELEVATE 在七个仿真任务序列中自主获取了墙面支撑抬升、支撑操作和墙面下降的控制器，并在后续到达与物体操作任务中复用。在初始控制器工作空间之外的五个任务上，ELEVATE 在五次独立运行中平均成功率达 98.4%-100%，而固定控制器库的编码智能体基线无法完成这些任务。合成的任务程序与获取的控制器无需真实机器人训练或微调即可迁移到物理四足操作机器人，在货架取物中达到 18/20 成功，在地面到货架放置中达到 16/20。

关键要点

01问题：任务规划智能体能力受限于现有控制器，无法完成超出工作空间的全身操作任务。
02方法：利用任务程序精炼中的持续失败触发编码智能体自动构建并精炼强化学习问题，获取缺失控制器。
03结果：在五个超出初始工作空间的任务上，五次运行平均成功率达 98.4%-100%，固定控制器库基线失败。
04结果：获取的控制器与任务程序零样本迁移到物理四足机器人，货架取物 18/20、地面到货架放置 16/20。
05局限：仅在七个仿真任务序列和有限物理试验中验证，未涉及真实机器人训练或微调。

解读

尚无解读。

原始英文摘要

arXiv:2610.38388v1 Announce Type: new Abstract: Quadrupedal manipulators can extend their workspace through whole-body interaction, but task-planning agents' capabilities remain limited by their available controllers. We present ELEVATE, a framework that uses persistent failures during task-program refinement to identify and acquire missing whole-body control capabilities. When these failures suggest a missing physical capability, a coding agent formulates and iteratively refines a reinforcement-learning problem, specifying behavior-specific rewards, curricula, and termination conditions. Validated controllers are added to the capability library and composed with existing controllers to retry the task. Starting from a single pretrained loco-manipulation controller, ELEVATE autonomously acquires controllers for wall-supported elevation, supported manipulation, and wall descent across a sequence of seven simulation tasks. These controllers are reused in subsequent reaching and object-manipulation tasks. On the five tasks beyond the initial controller's workspace, ELEVATE achieves mean success rates of 98.4%-100% across five independent runs, while coding-agent baselines with a fixed controller library fail to complete these tasks. The synthesized task programs, together with the acquired controllers, also transfer to a physical quadrupedal manipulator without real-robot training or fine-tuning, achieving 18/20 successes in rack retrieval and 16/20 in floor-to-rack placement. Videos and demonstrations are available at https://agentic-quad-elevate.github.io/public/.