SE3 Labs Releases A Wider Dream Eval, Opens 500-Episode Robot Demo Dataset on Hugging Face
paigeinsf · x · 2026-11-19
SE3 Labs published its 'A Wider Dream' evaluation and open-sourced the data: after finding Dream's zero-shot skills clustered in the table's middle third (where teleop data is densest), the team fine-tuned Dream and π0.5 on demos from a station with fully randomized layouts. Both models more than doubled their scores and spread skills across the whole table, though neither finishes tasks reliably yet — the next round will use DAgger.
The demonstrations are on Hugging Face: 500 episodes, 10 tasks, LeRobot format, CC BY 4.0.
More from Embodied
- LLMs zero-shot beat fine-tuned world action models on 7 of 10 robot tasks — paigeinsf · 2026-11-19
- Memory and Reasoning Remain LLM Territory: Fable Beats Dream-1 64 to 9 on Cup Task — paigeinsf · 2026-11-19
- The Secret Wasn't a Bigger Model: 500 Randomized Demos Lift Dream-1 Score 159% — paigeinsf · 2026-11-19
- Fine-tuned World Action Model Dream-1 Jumps 159%, Beats Frontier LLMs on Robot Tasks — paigeinsf · 2026-11-19
- Fine-tuned Dream-1 world action model jumps 159% on real-robot benchmark at 1/100th the cost — paigeinsf · 2026-11-19
- Robotics founder: "sim2real is BS" — labs buying data is a flash in the pan — Scobleizer · 2026-11-19