Robot training performance plateaus after turn 85
tristanbob · x · 2026-08-30
The author used Weights & Biases to visualize robot training data, revealing that the robot climbed quickly until turn 85, followed by a very slow incline. The author is investigating what caused this shift in performance.
Related event: RL-Trained Virtual Duck Fails to Walk Reliably After 5,000 Iterations(6 posts)→
More from Embodied
- Scoble heads to SF humanoid robotics conference, says absent Optimus is the elephant in the room — Scobleizer · 2026-10-17
- TANDEM mixes autonomous TAMP with on-demand teleoperation for 2.9× more demos — tomssilver · 2026-10-17
- DATAFARM gains a human-in-the-loop to handle what TAMP alone cannot — tomssilver · 2026-10-17
- Tesla supervised FSD approved in Slovakia, rollout begins soon — elonmusk · 2026-10-17
- DSP turns 15 minutes of human play into robot dexterity, no demos or sim2real needed — anand_bhattad · 2026-10-17
- NEURA shares full video and blog for Neuron AI ahead of GTC Berlin laundry demo — CyberRobooo · 2026-10-17