so100-pi05-new60-ft · step 76000

Pi0.5 policy for SO-ARM100 / SO100Grab the white cube to the white cup

  • 数据: tonghuiwang123/40 60 episodes / 13454 frames / 2×1280x720 @30fps
  • Checkpoint: step 7600045.2 epoch(有效 batch 8 = 4×2 DDP)
  • 起点: 旧60集(旧相机)训练的 pi0560 权重(两阶段)
  • 配置: dtype=bfloat16, train_expert_only=true, gradient_checkpointing=true
  • 可训练参数 693M / 总 4.14B(VLM 冻结)

相机命名(Pi0.5 专用,与 ACT/SmolVLA 不同)

物理位置 设备 本模型要求的 key
顶部 (top-down) /dev/video2 base_0_rgb
腕部 (wrist) /dev/video4 left_wrist_0_rgb

部署

lerobot-record \
  --robot.type=so100_follower --robot.port=/dev/ttyACM0 \
  --robot.id=my_awesome_follower_arm \
  --robot.cameras="{
                     base_0_rgb: {type: opencv, index_or_path: /dev/video2, width: 1280, height: 720, fps: 30},
                     left_wrist_0_rgb: {type: opencv, index_or_path: /dev/video4, width: 1280, height: 720, fps: 30},
                    }" \
  --dataset.repo_id=tonghuiwang123/eval_so100-pi05-new60-ft \
  --dataset.single_task="Grab the white cube to the white cup" \
  --policy.path=tonghuiwang123/so100-pi05-new60-ft

分辨率必须 1280x720。推理显存约 9.9 GB,需 12G 以上显卡。

Downloads last month
-
Safetensors
Model size
4B params
Tensor type
F32
·
BF16
·
Video Preview
loading