Bulk material handling is a critical, labor-intensive operation across various industries, traditionally performed by human operators using heavy hydraulic manipulators equipped with free-swinging, underactuated grippers. This work presents the first complete autonomous material handling solution deployed on a real-world 40-ton material handler.
Our system operates through a continuous grasp-and-dump cycle utilizing four key technical modules: a perceptive reinforcement learning (RL) attack-point planner, an obstacle-aware RRT* path planner, an RL waypoint-following controller, and an RL throwing controller that safely exploits passive pendulum dynamics for precise releases.
Across bulk pile management and constrained dump truck loading tasks, our framework matches expert human operators in scooped volume, achieves more compact material piles, and operates safely at a similar throughput as a moderately experienced human operator!
Read our full paper for a detailed description of the system and an in-depth analysis of the results!
doi.org/10.1109/TFR.2026.3662619
arxiv.org/abs/2508.09003
Authors: Filippo A. Spinelli, Yifan Zhai, Fang Nan, Pascal Egli, Julian Nubert, Thilo Bleumer, Lukas Miller, Ferdinand Hofmann, Marco Hutter
Work in collaboration with Liebherr-Hydraulikbagger GmbH, and partially supported by NCCR Digital Fabrication.
