Humanoid launches KinetIQ Ascend reinforcement learning for humanoid robot dexterity

London-based Humanoid introduced KinetIQ Ascend, a reinforcement learning method that refines robot manipulation skills through trial-and-error on industrial tasks. The approach builds a "capability factory" by starting with basic behaviors and letting RL improve them over days rather than months of manual tuning. Tests showed throughput gains of 42% to more than double across bin picking, tote handling, and bimanual lifting, with success rates reaching 98-99%. Humanoid, founded in 2024, has more than 250 engineers and partners with Bosch and Schaeffler. It aims to become the leading general-purpose industrial humanoid robotics company within two years.