en.Wedoany.com Reported - UK robotics and artificial intelligence company Humanoid has launched KinetIQ Ascend, a reinforcement learning-based technology solution aimed at achieving 99.9% operational reliability at human speed or faster.
Built on the previously released KinetIQ platform, KinetIQ Ascend directly improves robot performance in industrial tasks through trial-and-error learning. The system has been tested on multiple tasks, including picking parts from bins, handing objects to humans, and lifting and moving containers with both arms, proving effective across different operational scenarios.
In a machine loading application where robots pick steel bearing rings from bins and place them onto conveyor belts, KinetIQ Ascend increased throughput by 42%, enabling the robot to operate at 1.5 times the speed of the initially learned human demonstration. In another task involving picking items from cluttered totes and handing them to humans, the method boosted throughput by 85%, raising the success rate from 80% to 98%.
In a third dual-arm tote handling task, where the robot lifted totes from a table using both arms, throughput more than doubled, the success rate rose from 78% to 99%, and the failure rate decreased by approximately 20 times—all achieved after just a few days of training.
These results demonstrate a new way for KinetIQ Ascend to develop robotic capabilities, proven effective across a range of real-world operational tasks, from high-speed single-arm picking to complex dual-arm handling. The technology also shows that robot performance improves predictably with increased training time, similar to how large language models improve with more computation and data. Observed scaling trends, supported by simulation experiments, suggest that the approach can scale all the way to 100% reliability.
The new method also revealed two additional findings: improving only the most difficult parts of a workflow can enhance the entire task, and robots can generalize to objects not seen during training.
Humanoid Chief Technology Officer Jarad Cannon stated that the race for humanoid robots is becoming a matter of scale, and real-world reinforcement learning can be a core part of the solution. Robots that once required months of manual tuning can now surpass human demonstration performance within days. KinetIQ Ascend offers a new pathway for developing robotic capabilities, eliminating the need to spend months collecting data and manually tuning each new skill. Starting from basic behaviors, reinforcement learning refines them into deployable capabilities—a process termed building a "capability factory," marking the transition of humanoid robots from demonstrations to reliable industrial tools.
Humanoid outlined these findings in a new technical report covering the complete methodology of KinetIQ Ascend, including training infrastructure, algorithmic solutions, and a deeper analysis of the results.
