arXiv AI
StepTrigger: Contact-State-Triggered Backdoor Attacks on VLM-Powered Legged Robots
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Aug 7
Hijacking Robots with a Piece of Paper: A Systematic Study of Physical Prompt Injection in VLM-Controlled Robots
arXiv:2608. 05715v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) are increasingly deployed as planners in robotic systems, where they translate natural-language commands into executable actions grounded in visual scene understanding.
arXiv AI
Jul 7
!Imperio, smolVLA: The Implications of Data Poisoning on Open Source Robotics
arXiv:2607. 04146v1 Announce Type: cross Abstract: This work establishes that trigger-word data poisoning of vision language action models is practical, while at the same time the open-source robotics ecosystem holds trust assumptions about community contributions.
arXiv Computer Vision
Aug 28
TrapVLA: Trapping Vision-Language-Action Models in Configured Failure Modes
The paper introduces Configured Failure Trapping, a new backdoor attack targeting Vision‑Language‑Action (VLA) models that activates through subtle textual triggers and forces the robot to fail in a specific, controlled manner. It presents a data engine for generating high‑quality target trajectories, an automated evaluation suite, and two benchmarks—Trap‑LIBERO and Trap‑RoboTwin—covering four failure modes. The authors propose TrapVLA, a method that learns trigger‑induced action residuals to steer policies toward the desired failure behavior, demonstrating effectiveness in both simulation and real‑world robotic experiments while maintaining performance on clean data.
arXiv Machine Learning
Oct 22
Watch Your Step: Anticipated Perception for Humanoid Locomotion
The paper proposes a lightweight foothold prediction model (FPM) that anticipates future foot placements using only proprioception and velocity commands, allowing the robot to sample elevation maps only around predicted footholds. This preprocessing reduces redundant sensory input, enabling the locomotion policy to learn faster—reaching a given terrain difficulty in about one‑third the training iterations compared to using the full elevation map. The resulting policy outperforms both full‑map and attention‑based encoders on sparse foothold scenarios, and it zero‑shot transfers to a Unitree G1 humanoid across diverse indoor and outdoor obstacles.
arXiv AI
Jun 2
SilentDrift: Exploiting Action Chunking for Stealthy Backdoor Attacks on Vision-Language-Action Models
arXiv:2601. 14323v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models are increasingly deployed in safety-critical robotic applications, yet their security vulnerabilities remain underexplored.
arXiv Machine Learning
Jun 9
State Backdoor: Towards Stealthy Real-world Poisoning Attack on Vision-Language-Action Model in State Space
arXiv:2601. 04266v2 Announce Type: replace-cross Abstract: Vision-Language-Action (VLA) models are widely deployed in safety-critical embodied AI applications such as robotics.