Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Proactive Safe Human-Robot Collaborations via Data-Efficient Conditional Behavior Prediction

Ravi Pandya, Zhuoyuan Wang, Yorie Nakahira, Changliu Liu

TL;DR

This work addresses proactive, safe human–robot collaboration under data scarcity by formulating a model-based CBP framework that reasons about how a human will respond to the robot's future plan. It decouples the human's prior and post interaction goals and uses a Bayesian, softmax-based CBP to predict intent conditioned on the robot's plan, enabling proactive versus courteous behavior. A switching controller toggles between influence and courtesy based on detected human uncertainty, and a long-horizon safe controller samples from the CBP distribution to guarantee probabilistic safety. Across simulations and a user study, the approach demonstrates data-efficient performance comparable to learning-based baselines, improves interaction efficiency and perceived collaboration quality, and maintains high safety, with potential for online adaptation in future work.

Abstract

We focus on the problem of how we can enable a robot to collaborate seamlessly with a human partner, specifically in scenarios where preexisting data is sparse. Much prior work in human-robot collaboration uses observational models of humans (i.e. models that treat the robot purely as an observer) to choose the robot's behavior, but such models do not account for the influence the robot has on the human's actions, which may lead to inefficient interactions. We instead formulate the problem of optimally choosing a collaborative robot's behavior based on a conditional model of the human that depends on the robot's future behavior. First, we propose a novel model-based formulation of conditional behavior prediction that allows the robot to infer the human's intentions based on its future plan in data-sparse environments. We then show how to utilize a conditional model for proactive goal selection and safe trajectory generation around human collaborators. Finally, we use our proposed proactive controller in a collaborative task with real users to show that it can improve users' interactions with a robot collaborator quantitatively and qualitatively.

Towards Proactive Safe Human-Robot Collaborations via Data-Efficient Conditional Behavior Prediction

TL;DR

This work addresses proactive, safe human–robot collaboration under data scarcity by formulating a model-based CBP framework that reasons about how a human will respond to the robot's future plan. It decouples the human's prior and post interaction goals and uses a Bayesian, softmax-based CBP to predict intent conditioned on the robot's plan, enabling proactive versus courteous behavior. A switching controller toggles between influence and courtesy based on detected human uncertainty, and a long-horizon safe controller samples from the CBP distribution to guarantee probabilistic safety. Across simulations and a user study, the approach demonstrates data-efficient performance comparable to learning-based baselines, improves interaction efficiency and perceived collaboration quality, and maintains high safety, with potential for online adaptation in future work.

Abstract

We focus on the problem of how we can enable a robot to collaborate seamlessly with a human partner, specifically in scenarios where preexisting data is sparse. Much prior work in human-robot collaboration uses observational models of humans (i.e. models that treat the robot purely as an observer) to choose the robot's behavior, but such models do not account for the influence the robot has on the human's actions, which may lead to inefficient interactions. We instead formulate the problem of optimally choosing a collaborative robot's behavior based on a conditional model of the human that depends on the robot's future behavior. First, we propose a novel model-based formulation of conditional behavior prediction that allows the robot to infer the human's intentions based on its future plan in data-sparse environments. We then show how to utilize a conditional model for proactive goal selection and safe trajectory generation around human collaborators. Finally, we use our proposed proactive controller in a collaborative task with real users to show that it can improve users' interactions with a robot collaborator quantitatively and qualitatively.
Paper Structure (17 sections, 29 equations, 5 figures, 5 tables)

This paper contains 17 sections, 29 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Model-Based CBP
  5. The Robot's Multi-Stage Objective
  6. Safe Trajectory Generation
  7. Simulation Environment
  8. Score Function
  9. Simulated Human Models
  10. Baseline Robot Controllers
  11. Simulation Results
  12. User Study
  13. Conclusion and Future Work
  14. Human's Bayesian Goal Prediction
  15. Learning-based CBP Details
  16. ...and 2 more sections

Figures (5)

  • Figure 1: Overall diagram of the proposed framework, including the CBP model (Sec. \ref{['sec:cbp_model']}), courtesy and influence behavior (Sec. \ref{['sec:proposed_controller']}) and long-term safe control (Sec. \ref{['sec:safety']}).
  • Figure 2: Left: human chooses left goal in isolation. Right: robot chooses a goal to successfully influence the human's goal selection by modeling their goal change using model-based CBP.
  • Figure 3: Visualization of interaction and goal selection with (a) KL-divergence cost function (b) argmax belief cost function. The KL-divergence cost results in chattering of both agents' goals while the argmax belief results in a stable interaction.
  • Figure 4: Long-term safety probabilities in simulation.
  • Figure 5: Shows the setup for the user study where participants control an on-screen avatar directly with their hand to try and collect diamonds in collaboration with different robots.

Theorems & Definitions (1)

  • Remark III.1