Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multimodal Safe Control for Human-Robot Interaction

Ravi Pandya, Tianhao Wei, Changliu Liu

TL;DR

This work tackles safety for autonomous systems operating under multimodal uncertainty, with a focus on human-robot interaction where the human’s true intent is uncertain. It introduces the Multimodal Safe Set Algorithm (MMSSA), comprising Naive (N-MMSSA) and Optimal (O-MMSSA) variants that derive per-mode safety constraints from an energy-based safety index $\phi$, and enforce a multimodal chance constraint through $k$-sigma bounds. By optimizing per-mode safety parameters under a probabilistic constraint, O-MMSSA achieves a least-conservative safe control set while maintaining high safety, outperforming unimodal baselines (SEA) in multimodal settings. The approach is demonstrated in simulations of goaI-seeking and human-following robots with probabilistic human intent, showing improved safety performance without sacrificing efficiency, highlighting practical potential for safer HRI in uncertain environments.

Abstract

Generating safe behaviors for autonomous systems is important as they continue to be deployed in the real world, especially around people. In this work, we focus on developing a novel safe controller for systems where there are multiple sources of uncertainty. We formulate a novel multimodal safe control method, called the Multimodal Safe Set Algorithm (MMSSA) for the case where the agent has uncertainty over which discrete mode the system is in, and each mode itself contains additional uncertainty. To our knowledge, this is the first energy-function-based safe control method applied to systems with multimodal uncertainty. We apply our controller to a simulated human-robot interaction where the robot is uncertain of the human's true intention and each potential intention has its own additional uncertainty associated with it, since the human is not a perfectly rational actor. We compare our proposed safe controller to existing safe control methods and find that it does not impede the system performance (i.e. efficiency) while also improving the safety of the system.

Multimodal Safe Control for Human-Robot Interaction

TL;DR

This work tackles safety for autonomous systems operating under multimodal uncertainty, with a focus on human-robot interaction where the human’s true intent is uncertain. It introduces the Multimodal Safe Set Algorithm (MMSSA), comprising Naive (N-MMSSA) and Optimal (O-MMSSA) variants that derive per-mode safety constraints from an energy-based safety index , and enforce a multimodal chance constraint through -sigma bounds. By optimizing per-mode safety parameters under a probabilistic constraint, O-MMSSA achieves a least-conservative safe control set while maintaining high safety, outperforming unimodal baselines (SEA) in multimodal settings. The approach is demonstrated in simulations of goaI-seeking and human-following robots with probabilistic human intent, showing improved safety performance without sacrificing efficiency, highlighting practical potential for safer HRI in uncertain environments.

Abstract

Generating safe behaviors for autonomous systems is important as they continue to be deployed in the real world, especially around people. In this work, we focus on developing a novel safe controller for systems where there are multiple sources of uncertainty. We formulate a novel multimodal safe control method, called the Multimodal Safe Set Algorithm (MMSSA) for the case where the agent has uncertainty over which discrete mode the system is in, and each mode itself contains additional uncertainty. To our knowledge, this is the first energy-function-based safe control method applied to systems with multimodal uncertainty. We apply our controller to a simulated human-robot interaction where the robot is uncertain of the human's true intention and each potential intention has its own additional uncertainty associated with it, since the human is not a perfectly rational actor. We compare our proposed safe controller to existing safe control methods and find that it does not impede the system performance (i.e. efficiency) while also improving the safety of the system.
Paper Structure (20 sections, 32 equations, 4 figures, 2 tables)

This paper contains 20 sections, 32 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Energy-Function-Based Safe Control
  4. Multimodal Dynamical System
  5. Multimodal Safe Set Algorithm
  6. Naive Multimodal Safe Control (N-MMSSA)
  7. Optimal Multimodal Safe Control (O-MMSSA)
  8. Comparison with Unimodal Safe Control (SEA)
  9. Application to Human-Robot System
  10. Dynamics
  11. Human Intent Inference
  12. Simulation Setup
  13. Simulated Dynamics Models
  14. Goal Inference on Simulated Human
  15. Nominal Robot Control
  16. ...and 5 more sections

Figures (4)

  • Figure 1: Depiction of a multimodal human-robot system where the robot needs to stay safe with all potential goals $\theta_i$ the human might choose. The shaded region shows how safe potential future positions are, colored by the distance to the safe set of the corresponding control action.
  • Figure 2: The control constraints from each safe control method at the same system state.
  • Figure 3: Bayesian inference of the simulated human's goal
  • Figure 4: An example of unimodal safe control (Sec. \ref{['sec:sea']}) violating the minimum safe distance from the human while the proposed multimodal safe controller (Sec. \ref{['sec:o-mmssa']}) does not.