VLA Policies Do Not Know When Not to Act: Unintended Task Execution and Start Suppression
VLA Policies Do Not Know When Not to Act: Unintended Task Execution and Start Suppression
This research investigates the behavior of vision-language-action (VLA) policies when given missing, meaningless, or incorrect instructions. Using six model checkpoints in the LIBERO-Goal simulation environment, the authors found that robots consistently initiated movement even with empty inputs and frequently completed unspecified tasks. While a gating mechanism successfully suppressed motion on non-task inputs for five of the six models, fine-tuning approaches failed to reliably keep robots still. The study concludes that current VLA architectures lack the ability to distinguish between valid commands and noise, posing potential risks in real-world deployment where input reliability cannot be guaranteed.
The study demonstrates that current vision-language-action (VLA) policies exhibit unintended task execution and motion initiation when provided with empty, nonsense, or infeasible instructions, highlighting a critical safety gap in autonomous decision-making under ambiguous input conditions.
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.
- Environment
- SimulationOpen source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
d protocol with empty, neutral, nonsense, infeasible, and wrong-task instructions to six model checkpoints in LIBERO-Goal simulation. Holding the initial state fixed and varying only the instruction, we ask three questions: whether the robot starts to move, which task it completes, and what the instruction determines. Even when the instruction was empty, the robot started moving in every episode with all six checkpoi
Open source S3
nts, and with five of them it often completed a task that the instruction had not specified. On two GR00T checkpoints, the instruction determined which task was executed but not whether the robot started moving; the robot moved even on infeasible instructions. We then tested three methods for suppressing the start: prepending "if no task is given do not move" to the instruction; a gate that holds the robot when the f
Open source S4
irst action chunk for the given instruction nearly equals that for an empty instruction; and fine-tuning the policy to hold still on non-task inputs. The prefix stopped the robot in none of the six checkpoints, and the gate held it on non-task inputs in five of the six checkpoints. The fine-tuned policy never completed a task on non-task inputs, but it did not reliably keep the robot still, and it failed to suppress
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org