SafetyALFRED: Evaluating Safety-Conscious Planning of Multimodal Large Language Models
Josue Torres-Fonseca, Naihao Deng, Yinpei Dai, Shane Storks, Yichi Zhang, Rada Mihalcea, Casey Kennington, Joyce Chai
THE PROBLEM
This paper focuses on Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Control & PlanningPlanningFiguring out what the robot should do before or during movement.. This paper exposes a critical gap in Modern Robot LearningMultimodalUsing more than one type of input, like vision, language, touch, or proprioception. LLMs: they can recognize kitchen hazards in text-based QA but fail to plan safe mitigation actions in embodied environments. If you're building Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. agents with vision-language models, you need to know this—your model might verbally identify a spill but physically walk into it anyway. Read the paper by tracking the Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. definition, the Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. or data assumptions, and the evidence that supports the claimed improvement.
HOW IT WORKS
Task framing
Core method
Data and supervision
Evaluation evidence
KEY RESULTS
This paper exposes a critical gap in Modern Robot LearningMultimodalUsing more than one type of input, like vision, language, touch, or proprioception. LLMs: they can recognize kitchen hazards in text-based QA but fail to plan safe mitigation actions in embodied environments. If you're building Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. agents with vision-language models, you need to know this—your model might verbally identify a spill but physically walk into it anyway.
WHY DEVELOPERS SHOULD CARE
This paper exposes a critical gap in Modern Robot LearningMultimodalUsing more than one type of input, like vision, language, touch, or proprioception. LLMs: they can recognize kitchen hazards in text-based QA but fail to plan safe mitigation actions in embodied environments. If you're building Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. agents with vision-language models, you need to know this—your model might verbally identify a spill but physically walk into it anyway.
LIMITATIONS
The main limitation to check is whether the claimed behavior holds outside the paper's reported setup. That means testing across different Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. embodiments, scenes, objects, and data distributions.
WHAT COMES NEXT
The practical next step is independent reproduction with clear baselines, ablations, and stress tests. For a developer, the useful follow-up is to map the paper's Core ConceptsTaskThe job the robot is supposed to complete, such as pick-and-place, navigation, or drawer opening. Control & PlanningPlanningFiguring out what the robot should do before or during movement. assumptions onto a concrete Core ConceptsRobotA physical system with sensors and actuators that can observe the world and take actions. stack, then test the smallest version of the method that could run end to end.