dp-cupbasket-abl-visual

LeRobot diffusion policy (ROBOTIS lerobot-cyclo fork, 288 M parameters) for the bimanual cup handover onto a basket with the ROBOTIS FFW SG2 (SG2 + CAD grippers, calibrated wrist cameras), cup DR ablation (visual).

Files at the repo root = step 100k (DiffusionPolicy.from_pretrained("padiredd/dp-cupbasket-abl-visual")); step-10000/, step-20000/, step-30000/, step-40000/, step-50000/, step-60000/, step-70000/, step-80000/, step-90000/ = intermediate checkpoints (every 10k steps).

data padiredd/Ablation_Cup_visual: 2,000 successful sim demos from the cup-handover visual vs physical randomization ablation (data_collection/cup_basket/ablation_vp, 2026-11-07): the v6 program cup_handover_v6.py (right close recorded 0.70 / executed 0.50), seeds 60001+, --split_rng (layouts, program params and gripper values identical across the four sets per pair_key) and cup-safe gripper randomization in all four (open 0.10-0.157 both arms, left executed close 1.3-1.5), cut at exactly 2,000 successes in part order. Trained with the same frozen recipe as the block-pick ablation DPs (padiredd/dp-blockpick-cap2k-*-dr). This set: visual randomization only: table look, backdrop, cup colour jitter, dr2 room lights, spot lights over the work area
sampling uniform over all frames (LeRobot default EpisodeAwareSampler)
inputs / outputs head + left wrist + right wrist cameras, each 240x240, + 22-dim joint state; action = 16 joint targets, 15 Hz
image augmentation heavy, random order, up to 5 of: brightness 0.6-1.4, contrast 0.6-1.4, saturation 0.3-1.7, hue +-0.1, sharpness 0.3-2.0, affine +-10 deg / 10 % / 0.9-1.1, perspective 0.2, blur, random erasing, grayscale
state noise none
recipe 100k steps, batch 32, ResNet18 per camera, Adam 1e-4, cosine, DDPM 100 steps, horizon 64 / 32 executed, n_obs_steps 1, seed 1000
hardware one RTX 4090 24 GB (lab225), shared with at most one other ablation run

Task string: "pick up the blue cup with the left hand, hand it to the right hand, and place it on the basket". Evaluation results are added once they are run.

Downloads last month
22
Safetensors
Model size
0.3B params
Tensor type
F32
·
Video Preview
loading

Collection including padiredd/dp-cupbasket-abl-visual