Towards Safe Robot Foundation Models Using Inductive Biases

2025-05-15

AI Summary

While existing robotic foundation models exhibit strong generalization capabilities, they lack formal safety guarantees; behavioral cloning fails to provide rigorous safety assurances and relies heavily on large-scale safe demonstration data. Method: We propose the first approach to embed formal safety constraints directly into the action output layer of a general-purpose robotic foundation model. Our method integrates geometric priors with ATACOM to construct an online safety layer that projects actions onto a safe subspace satisfying kinematic, dynamic, and task-specific constraints—including collision avoidance, joint limits, and high-speed trajectory feasibility. Contribution/Results: This framework decouples safety enforcement from task generalization, requiring neither safety-specific fine-tuning nor additional safety-labeled data. Evaluated on real-world tasks—including robotic grasping and robot hockey—it achieves zero collisions and full constraint satisfaction while preserving task performance, thereby demonstrating synergistic enhancement of formal safety and generalization capability.

Technology Category

• Natural Language Processing: Safety and Robustness
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness
• Intelligent Robots: Other Foundations and Applications

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Security and privacy of machine learning and AI applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Safety is a critical requirement for the real-world deployment of robotic systems. Unfortunately, while current robot foundation models show promising generalization capabilities across a wide variety of tasks, they fail to address safety, an important aspect for ensuring long-term operation. Current robot foundation models assume that safe behavior should emerge by learning from a sufficiently large dataset of demonstrations. However, this approach has two clear major drawbacks. Firstly, there are no formal safety guarantees for a behavior cloning policy trained using supervised learning. Secondly, without explicit knowledge of any safety constraints, the policy may require an unreasonable number of additional demonstrations to even approximate the desired constrained behavior. To solve these key issues, we show how we can instead combine robot foundation models with geometric inductive biases using ATACOM, a safety layer placed after the foundation policy that ensures safe state transitions by enforcing action constraints. With this approach, we can ensure formal safety guarantees for generalist policies without providing extensive demonstrations of safe behavior, and without requiring any specific fine-tuning for safety. Our experiments show that our approach can be beneficial both for classical manipulation tasks, where we avoid unwanted collisions with irrelevant objects, and for dynamic tasks, such as the robot air hockey environment, where we can generate fast trajectories respecting complex tasks and joint space constraints.

Problem

• Ensuring safety in robot foundation models without formal guarantees
• Addressing lack of safety constraints in current robot learning approaches
• Combining foundation models with geometric biases for safe state transitions

Innovation

• Combining robot foundation models with geometric inductive biases
• Using ATACOM safety layer for safe state transitions
• Ensuring formal safety guarantees without extensive demonstrations