Sayood/tanitad-rollout-recovery

TanitAD — rollout-recovery arms (RR-20 / RR-CTL)

Weights from the rollout-recovery experiment, pushed 2026-08-02. These existed on a single pod disk and nowhere else.

The experiment

Both arms fine-tune v1 (flagship4b-speedjerk-30k) for 2,000 steps to step 31999 — same pod, same seed, exactly one flag different.

	RR-CTL (--rollout-k 4)	RR-20 (--rollout-k 20)
ade@2s	0.424 [0.368, 0.483]	0.348 [0.291, 0.411]
paired ΔCI	[0.0613, 0.0906] — separated, RR-20 wins
speed bias	+0.9397 m/s	⭐ -0.0092 m/s
speed MAE	2.3122	1.6965
along-track final bias	+0.3355 m	-0.0484 m
curvature MAE	0.021798 1/m	⚠️ 0.048254 1/m
yaw-rate MAE	8.848 °/s	9.8445 °/s
miss@2m	0.043	0.056

n = 881 windows over the canonical 40 validation episodes; paired episode-cluster bootstrap, B = 2000.

What it means

Training on longer recursive rollouts essentially ERASED the longitudinal speed bias (+0.94 → -0.009 m/s). That bias is the programme's largest measured defect — an oracle target-speed recovers 88.7 % of the goal-conditioning gap, all of it longitudinal.

⚠️ It was paid for in lateral fidelity: curvature error 2.2× worse, yaw-rate worse, and miss@2m rose while ADE fell. On ADE alone this reads as a clean win; it is a trade, visible only because the evaluation reports longitudinal, lateral, tactical and strategic families separately rather than a single scalar.

⛔ RR-20 must be compared to RR-CTL, never to v1. Against v1 you would confound rollout-k with 2,000 extra fine-tuning steps.

Contents

• rr20/model_grounding.pt — RR-20: model + grounding + step (optimizer state dropped)
• rrctl/model_grounding.pt — RR-CTL, same structure
• refc-base-e1f-junction/ — a REF-C junction arm, also single-disk
• provenance/ — run status files

Trained on NVIDIA PhysicalAI-AV. Architecture flagship4b, 276.9 M params, speed_input on (action_dim 3).