Humanoid says KinetIQ Ascend reinforcement learning approaches human-level dexterity

A London-based robotics company has introduced a reinforcement learning system designed to improve robot manipulation tasks to near-human reliability levels. The system works by allowing robots to learn and refine skills directly through trial-and-error on industrial tasks rather than requiring extensive manual programming and data collection. In testing across picking, handling, and bimanual tasks, the approach reportedly achieved significant improvements in speed and success rates within days of training. The company claims this method scales predictably with more training time, similar to how large language models improve with additional compute and data.