← Leaderboard

VPP (Video Prediction Policy)

World-action
Tsinghua University & RobotEra · Dec 19, 2024

Generalist policy that conditions actions on the internal representations of a video-prediction model fine-tuned on robot and human manipulation video. It reports CALVIN, plus self-reported real-robot results (0.85 seen / 0.73 unseen on Franka). It is the predecessor of RobotEra’s VPP2.

Type
World-action
Parameters
—
Backbone
Stable Video Diffusion 1.5B + diffusion-transformer action head
Weights
Open
License
MIT
Commercial use
Allowed
Released
Dec 19, 2024
Robots
Franka Panda, XArm with XHand, Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
CALVIN ABC→DNear saturationSimulation
#11 of 244.33 / 5, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error