Hierarchical RL Framework for Legged Manipulator Pick-and-Place
A team of researchers has created a hierarchical reinforcement learning system designed for dynamic pick-and-place operations utilizing a quadruped robot fitted with a 6-DOF arm. This system features a dedicated mass estimation module that allows for adaptive control of the entire body when handling objects of different weights. In simulations, the framework recorded a success rate of 86.05% with loads reaching 2.3 kg. Additionally, real-world tests conducted in six varied scenarios, with controlled changes in object size, weight, and task heights, achieved an average success rate within a vertical range extending from ground level to tabletops at 1.1 meters high.
Key facts
- Hierarchical reinforcement learning framework for legged manipulator
- Quadruped with 6-DOF robotic arm
- Explicit mass estimation module for adaptive control
- 86.05% success rate in simulation with payloads up to 2.3 kg
- Real-world validation across six scenarios
- Vertical workspace from ground to 1.1 m high tabletops
- Controlled variations in object size, mass, and task heights
Entities
—