TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency

Key points

• We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation.1
• We introduce TacitVLA, a reasoning VLA whose action expert never attends to the reasoning it generates for the current call.1
• On long-horizon tasks where the reasoning text also serves as episodic memory, it can generate each entry during action execution rather than before acting.1
• We evaluate it in a controlled study on four benchmarks spanning manipulation and autonomous driving, memory and non-memory tasks: RoboMME, RMBench, LIBERO, and PhysicalAI-AV.1

Sources (2)

• [1]TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency
  Hugging Face Daily Papers · Oct 13, 12:00 AM

  Textual reasoning before acting has improved vision-language-action (VLA) models, but at the cost of latency: every action conditioned on the reasoning text must wait for autoregressive decoding.

  We show that actions do not need to wait for or condition on the generated reasoning to benefit from it, because the reasoning supervision already improves the backbone representation.

• [2]TacitVLA: Reasoning Vision-Language-Action Models Without Reasoning Latency
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 01:09 AM · same content

Extractive summary: sentences quoted from the sources.