LingBot-VA — Galaxea A1 Mango Placement EEF — Step 200

LingBot-VA fine-tuned for red-mango placement into a bowl or plate with a Galaxea A1 arm. The model predicts robot actions and video latents.

Data

Training demonstrations.

Field	Value
Episodes	52
Frames	17,081 at 30 FPS
Tasks	Red mango to blue plate; red mango to bowl
Cameras	`front` 480×480 RGB; `wrist` 640×480 RGB
Action	Episode-relative EEF pose + continuous normalized gripper

Training epochs

Approximately 61.54 equivalent fine-tuning epochs, calculated as 200 optimizer updates × 16 effective global batch size / 52 training segments. Each selected episode contributes one latent training segment; the denominator is the segment count, not the raw video-frame count. This estimate assumes all selected segments have valid latent caches and excludes base-model pretraining. See training_summary.json.

Evaluation

Real-robot rollouts and object-grounding analysis.

Files and configuration

transformer/ contains the step-200 weights. The base tokenizer, text encoder, and VAE are included. Use configs/va_a1_cfg.py for the EEF action mapping and normalization.

Training details · Training summary