Gen2Act C44 RoboLab task-holdout — epoch 10
EMA-only model parameters from epoch 10 of Gen2Act experiment C44, initialized from C39 and fine-tuned for 10 epochs on RoboLab small.
Five complete tasks were held out from training: BananasInBinThreeTotalTask, CookingClearPlateTask, FruitsOnPlate3Task, RubiksCubeTask, and StackWhiteMugsTask. The training set contained 38 other tasks (95 rollouts); validation contained 24 rollouts from the five held-out tasks.
Epoch-10 held-out-task metrics: action MAE 0.064795, action RMSE 0.096883, gripper accuracy 84.93%, validation loss 0.149419.
pytorch_model.pt is a PyTorch EMA state dict, not a standalone Transformers model. Load it using the included Gen2Act config.json.
- Downloads last month
- 11