Quadruped robots learn new whole-body skills from failure feedback
ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation
Robotics
Summary
Quadruped robots that can move and manipulate objects often need new skills to complete complicated tasks but don’t always know what skills they lack. The authors present ELEVATE, a system where the robot notices when it fails at a task, then automatically creates and improves new controllers to fix that problem using learning techniques. These new skills are added to the robot’s library, allowing it to solve harder problems. Their approach was tested in simulations and on a real robot without extra training, achieving high success on placing and retrieving objects.
What this means in practice
- •For robotics engineers: Create adaptive robot controllers that expand locomotion and manipulation abilities by learning from failures during tasks.
- •For warehouse automation teams: Deploy quadruped robots for rack retrieval and item placement using learned whole-body skills without additional robot retraining.$Commercial implications: Enables sale of more versatile quadruped robots capable of handling complex material handling tasks autonomously.
Authors
Shiqi Liu, Radu Corcodel, Ding Zhao
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/.