UK-based robotics company Humanoid has launched KinetIQ Ascend, a reinforcement learning (RL) system designed to achieve 99.9% manipulation reliability at human speed or faster. Building on the KinetIQ platform, Ascend uses trial-and-error learning to refine robot skills directly on industrial tasks, reducing the need for manual tuning and extensive data collection.
Key takeaways
- In a machine-feeding task (picking steel bearing rings from a bin to a conveyor), throughput rose 42%, with the robot operating at 1.5× the speed of human demonstrations.
- For picking items from a cluttered tote and handing to a person, throughput increased 85% and success rates improved from 80% to 98%.
- In a bimanual tote-handling task, throughput more than doubled, success rates rose from 78% to 99% (a ~20× reduction in failures) after only days of training.
- Performance scales predictably with training time, suggesting a path to 100% reliability, similar to scaling laws in large language models.
- Robots generalized to unseen objects, and improving only the hardest part of a workflow improved the entire task.
The results indicate that RL can transform humanoid robots from demo-stage to reliable industrial tools, with Humanoid's CTO describing the approach as a 'capability factory.' The company has published a technical report detailing the methodology and results.
Source: Robotics & Automation News (roboticsandautomationnews.com) · Published 2026-07-06 · “UK startup Humanoid launches reinforcement learning system to improve robot manipulation”