T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models
T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models
This paper introduces T3R, a training-free two-stage token refinement framework designed to enhance the efficiency and robustness of Vision-Language-Action (VLA) models.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
milliseconds per inference call attributed to pruning and attention bias
7.8
View original evidence
matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA ba
Open source S5speedup factor over unpruned baseline
3.3
View original evidence
matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA ba
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
attention biasing reallocates attention toward action-critical visual regions and language instructions within the retained tokens. Together, pruning produces a cleaner candidate set, while attention biasing further emphasizes the most decision-relevant evidence, yielding a better accuracy-efficiency frontier than either stage alone. On LIBERO-Spatial, T3R removes 85% of scene-camera visual tokens while essentially
Open source S4
matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA ba
Open source S5
ckbones show the strongest accuracy-efficiency benefits in multi-camera settings. These results demonstrate that training-free token refinement is a practical approach to efficient and robust VLA inference.
Open source S6
arXiv:2611.08413v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models enable end-to-end robot control, but inference is increasingly expensive. Each decision processes hundreds of visual tokens, often across multiple cameras, leading to substantial computation while allowing irrelevant background content to distract action prediction under distribution shift. We argue that these issues stem from two com
Open source S2
T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models
Open source S1
Source：arXiv Robotics — research abstracts · arxiv.org