Researchers release desktop robot platform for competitive learning experiments

Researchers have released KlaskTron, an open-source robotic testbed costing less than $2500 USD, and report transferring trained agents from simulation to physical play without additional training on the hardware. The preprint presents a desktop platform based on the tabletop game KLASK for studying how competing robots learn. Two motor-driven gantries control play. A digital replica in NVIDIA Isaac Lab and a learned model of the actuators support the transfer from simulation. The release includes hardware designs, parts lists, assembly documentation, inference software and the simulation setup. The demonstrated task is competitive tabletop play; the reported transfer does not establish performance on other robotic tasks.