From November 9th to 12th, the Robotics Learning Conference 2026 (CoRL 2026) was held in Austin, Texas, USA. Zeeshan Zia, a CoRL 2026 attendee, presented a slide showcasing top-tier laboratory research. The slide showed that in a fixed process of picking crankshafts one by one from a bin and placing them onto a conveyor belt, after approximately 15 weeks of fine-tuning using teleoperational intervention data, the latest model, LBM 2.0, achieved an average intervention interval (MTBI) of 340.3 seconds. This is approximately 4.3 times faster than the previous generation's 78.5 seconds and approximately 15 times faster than the first generation's 22.6 seconds. The slide also indicated an autonomy rate of 98.2%. Based on this, the research team estimated that 3 remote assistants could manage 50 robots. However, Zeeshan believes that real-world factory tasks are more dispersed, and the actual management ratio may be closer to 1 operator per 5 robots. Even in highly standardized and repetitive processes, robots still require human intervention on average every 5 minutes and 40 seconds, indicating that the ability to operate autonomously continuously remains a significant bottleneck for the commercialization of humanoid robots, and the frequency of human intervention is a key factor affecting large-scale deployment and operational economics.
