Disassembling high-voltage Electric Vehicle (EV) batteries is a dangerous, manual process that creates a bottleneck for the circular economy. To solve this challenge, the Eureka-funded Project LAMBDA brought together three partners—AICA, Circu Li-ion, and the Swiss Battery Technology Center (SBTC)—to teach an industrial robot how to perform this complex task with a human-like sense of touch.

The project’s final testing concluded at Circu Li-ion's battery disassembly line in Luxembourg. There, the team successfully deployed their software on an industrial Kuka robot, demonstrating a pioneering application of an AI-based tactile control policy in a production environment. The system used 3D vision and force feedback on a neural network trained in simulation with Reinforcement Learning (RL) to drive the robot, culminating in the successful cutting and removal of a cable harness on a real, high-voltage EV battery pack. This achievement represents a significant step in taking human operators out of harm's way and bringing learning-based manipulation to industrial applications.

This successful demonstration was the result of a deeply integrated partnership. The SBTC contributed the foundational research on robot learning and sim-to-real methods. They developed simulated environments and low-level control strategies for dexterous robotic skill training, and optimized RL workflows for training and deployment, reducing the sim-to-real gap in a controlled lab environment. The findings from their work were used to identify the best opportunities for using RL in industrial settings.