DEEP Robotics reports real-robot tests across embodied-AI research

What happened

In its October 24 announcement, the company says its DR-APO training method raised a LYNX M20 robot's navigation success from 25% to 65% in its real-world tests, while a separate simulated test rose from 70.33% to 89.55%. It also describes LynxScenes, a recorded outdoor dataset gathered with the LYNX M20 over 833.6 hours and 3,294.7 kilometers. The studies involve researchers at several Chinese and overseas universities.

Why it matters

The work links model training with physical robot experiments and a shared recorded-data benchmark. The numerical gains are company-reported results against selected baselines; they do not establish independent performance, a deployed service or reliable operation across all outdoor conditions.

Source checked October 31, 2026. Status describes the evidence in the linked source at the time of review.