OGBench Cube
A manipulation task environment in the OGBench benchmark suite involving cube manipulation, used to evaluate model-based reinforcement learning algorithms.
Timeline (last 90 days)
The study tests LeWorldModel (LeWM) on two environments, Reacher and OGBench Cube.
At the smallest perturbation scale, the dominant action direction of LeWorldModel deviates from the simulator's true direction: about 36 degrees on Reacher and about 42 degrees on Cube.
After the perturbation is magnified tenfold, the action geometry discrepancy of LeWorldModel shrinks significantly: by about 70% on Reacher and about 47% on Cube.
All Facts (3)
The study tests LeWorldModel (LeWM) on two environments, Reacher and OGBench Cube.
50%UnverifiedAt the smallest perturbation scale, the dominant action direction of LeWorldModel deviates from the simulator's true direction: about 36 degrees on Reacher and about 42 degrees on Cube.
50%UnverifiedAfter the perturbation is magnified tenfold, the action geometry discrepancy of LeWorldModel shrinks significantly: by about 70% on Reacher and about 47% on Cube.
50%