disentangled-vla/libero-island-ablation-v4-basevla-aug-p05

V4 BaseVLA + Domain Augmentation (p=0.5) — LIBERO Island Eval

Ablation V4 of the Stage3 Qwen2.5-VL framework: BaseVLA with image-level domain augmentation (lighting + warp randomization, applied with probability p=0.5).

Evaluated on the LIBERO-Spatial island viewpoint protocol from Xing et al. 2025 (CoRL) "Shortcut Learning in Generalist Robot Policies".

Architecture

• —Backbone: Qwen/Qwen2.5-VL-3B-Instruct + LoRA (r=32, q/v_proj)
• —Single AttentiveLatentHead A (8 queries, depth 2, 16 heads, proj_dim=4096)
• —ResNetActionHead (hidden 2048, 2 blocks, chunk=16, action_dim=32 padded to 7-D)
• —Loss: L1(action_pred[..., :7], gt_action[..., :7]) only
• —Image-level augmentation during training (lighting 0.3–2.0, warp 0.08–0.20, applied with p=0.5)
• —No InfoNCE / no SIGReg / no V-JEPA target

Training

• —Data: LIBERO-Spatial island A+B paper-200 protocol (~211K frames, 1,724 eps)
• —bs=64 single GPU, lr=1e-4 cosine, 100K steps, 10 epochs

Evaluation protocol

6 settings × 5 tasks × 5 trials = 25 episodes / setting.

Results — epoch_10

Image-level augmentation closes most of the cross-domain gap (V1 OOD 0% → V4 OOD 54%), at slight cost on in-domain. A simple but strong baseline.

Files

eval_ablation_v4_first_videos/
├── {setting}_results.jsonl
├── {setting}_summary.json
└── videos/{setting}/task{N}_trial{T}_a{angle}_{OK,FAIL}.mp4

Videos: 25 ep × 6 settings = 150 mp4 total.