A CSDN write-up describes how Pigey pushed the π0.5 robot policy from a 16.7% real-robot success rate to 97.3% without training model weights. For teams working on embodied AI, the interesting part is not the headline number but the implied workflow: adaptation that avoids full retraining could cut iteration cost and deployment time significantly. That matters for startups and labs that cannot afford large-scale fine-tuning cycles for every new environment or task. The claim still needs independent reproduction, and the post is a single-source report rather than a peer-reviewed benchmark. Treat it as a directional signal: training-free or lightweight policy adaptation is becoming a competitive axis in robotics. Watch for follow-up evaluations, open-source releases, and whether similar gains hold across different robot platforms and task distributions.
A Chinese engineering post reports that Pigey raised the π0.5 robot policy's real-world success rate from 16.7% to 97.3% without training weights. If it holds up, this is a notable signal for cheaper embodied AI adaptation.