Imle-vla speeds robot action with single-step vision language model
IMLE-VLA: Fast Single-Step Action Generation for Vision-Language-Action Policies
RoboticsComputer Vision and Pattern Recognition
Summary
Robots that understand images and language often take many small steps to decide what to do next, making them slow and jerky. The authors propose IMLE-VLA, a method that generates robot actions in one quick step instead of multiple steps. This makes the robot move smoother and finish tasks faster without losing accuracy or flexibility. They tested IMLE-VLA on many tasks in simulation and on a real robot, showing it is faster and just as reliable as previous methods.
What this means in practice
- •For robotics engineers: Build robots that can decide actions much faster using single-step vision-language models for smoother and quicker task execution.
- •For industrial automation teams: Improve robotic arm efficiency on production lines by replacing slow iterative action decisions with fast single-step action generation.
Authors
Kian Hosseinkhani, Qinhe Peng, George Shramko, Mehran Aghabozorgi, Jianing Qian, Tristan Engst, Alireza Moazeni, Dinesh Jayaraman, Ke Li
Abstract
Vision-language-action (VLA) policies leverage pretrained vision-language backbones to achieve strong cross-task generalization. A leading design couples this backbone with a dedicated continuous action head trained via diffusion or flow matching. However, such heads rely on iterative multi-step sampling, for example 10 Euler steps in $π_{0.5}$. This creates an inference bottleneck that produces stop-and-go movement in the robot and slower task completion. We introduce IMLE-VLA, which replaces the iterative action head with a single-step conditional generator trained via conditional Implicit Maximum Likelihood Estimation (cIMLE). The cIMLE objective promotes multimodal action coverage, avoiding the mode collapse of naive regression heads while eliminating multi-step sampling entirely. When IMLE-VLA is applied to $π_{0.5}$, it increases inference frequency 3.67x (55 Hz vs. 15 Hz), enabling up to 11x higher action throughput. On the 40-task LIBERO benchmark, IMLE-VLA achieves the highest average success rate (98.0%) among all baselines while leading in inference frequency. Under the test-time perturbations of LIBERO-plus, IMLE-VLA retains $π_{0.5}$'s robustness while other baselines degrade sharply, confirming that the cIMLE head preserves generalization. Real-world experiments on a Franka Emika Panda across four tasks demonstrate smoother motion (2.2x to 3.0x lower jerk) and faster task completion, with IMLE-VLA outperforming $π_{0.5}$ on every task and reducing average VLA inference time per episode by 3.9x to 6.6x. Videos and code are available at https://kianhk6.github.io/IMLE-VLA/