Inference-time safeguards
Detect and regulate unsafe concepts while an embodied policy is deciding what action to take.
Research programme · 04
Embodied Intelligence: Trustworthy Multimodal Robotics
How can intelligent agents remain safe when perception becomes physical action?
Trust mechanisms for vision-language-action models and embodied agents, including inference-time safeguards, adversarial testing, and physical-world backdoors.
3 selected research outputs
Research focus
Three questions shape this programme.
This is the working agenda: a precise statement of the problems the lab is building toward, not a claim that every component is already solved.
Adversarial VLA evaluation
Expose model-agnostic failure modes across vision, language, and action interfaces.
Physical-world threat models
Study how ordinary-looking objects and environmental cues can trigger goal-directed unsafe behaviour.
Selected publications & outcomes
Evidence behind the direction.
Peer-reviewed papers and clearly labelled arXiv preprints are shown with their source records. See the complete Google Scholar profile ↗ for the full publication history.
Concept-Based Dictionary Learning for Inference-Time Safety in Vision Language Action Models
Uses concept-based dictionary learning to support inference-time safety in vision-language-action models.
Primary sourceModel-agnostic Adversarial Attack and Defense for Vision-Language-Action Models
Studies model-agnostic adversarial attacks and defences for vision-language-action systems.
Primary sourceGoal-oriented Backdoor Attack against Vision-Language-Action Models via Physical Objects
Examines goal-oriented backdoor attacks on vision-language-action models through physical objects.
Primary sourceThe full agenda
Explore all four research layers.
Students · collaborators · partners
Help move this programme from a hard question to reliable practice.
RISE-AI welcomes people who value rigorous foundations, reproducible systems, and generous teamwork.
See openings and how to apply →