Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Learning-Based Framework for Safe Human-Robot Collaboration with Multiple Backup Control Barrier Functions

Neil C. Janwani, Ersin Daş, Thomas Touma, Skylar X. Wei, Tamas G. Molnar, Joel W. Burdick

TL;DR

This paper develops a broadcast scheme that estimates driver intention and integrates BCBFs with multiple backup strategies for human-robot interaction and demonstrates the method’s efficacy on a dualtrack robot in obstacle avoidance scenarios.

Abstract

Ensuring robot safety in complex environments is a difficult task due to actuation limits, such as torque bounds. This paper presents a safety-critical control framework that leverages learning-based switching between multiple backup controllers to formally guarantee safety under bounded control inputs while satisfying driver intention. By leveraging backup controllers designed to uphold safety and input constraints, backup control barrier functions (BCBFs) construct implicitly defined control invariance sets via a feasible quadratic program (QP). However, BCBF performance largely depends on the design and conservativeness of the chosen backup controller, especially in our setting of human-driven vehicles in complex, e.g, off-road, conditions. While conservativeness can be reduced by using multiple backup controllers, determining when to switch is an open problem. Consequently, we develop a broadcast scheme that estimates driver intention and integrates BCBFs with multiple backup strategies for human-robot interaction. An LSTM classifier uses data inputs from the robot, human, and safety algorithms to continually choose a backup controller in real-time. We demonstrate our method's efficacy on a dual-track robot in obstacle avoidance scenarios. Our framework guarantees robot safety while adhering to driver intention.

A Learning-Based Framework for Safe Human-Robot Collaboration with Multiple Backup Control Barrier Functions

TL;DR

This paper develops a broadcast scheme that estimates driver intention and integrates BCBFs with multiple backup strategies for human-robot interaction and demonstrates the method’s efficacy on a dualtrack robot in obstacle avoidance scenarios.

Abstract

Ensuring robot safety in complex environments is a difficult task due to actuation limits, such as torque bounds. This paper presents a safety-critical control framework that leverages learning-based switching between multiple backup controllers to formally guarantee safety under bounded control inputs while satisfying driver intention. By leveraging backup controllers designed to uphold safety and input constraints, backup control barrier functions (BCBFs) construct implicitly defined control invariance sets via a feasible quadratic program (QP). However, BCBF performance largely depends on the design and conservativeness of the chosen backup controller, especially in our setting of human-driven vehicles in complex, e.g, off-road, conditions. While conservativeness can be reduced by using multiple backup controllers, determining when to switch is an open problem. Consequently, we develop a broadcast scheme that estimates driver intention and integrates BCBFs with multiple backup strategies for human-robot interaction. An LSTM classifier uses data inputs from the robot, human, and safety algorithms to continually choose a backup controller in real-time. We demonstrate our method's efficacy on a dual-track robot in obstacle avoidance scenarios. Our framework guarantees robot safety while adhering to driver intention.
Paper Structure (14 sections, 2 theorems, 19 equations, 5 figures)

This paper contains 14 sections, 2 theorems, 19 equations, 5 figures.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Control Barrier Functions
  4. Backup Control Barrier Functions
  5. Deep Neural Networks
  6. Framework
  7. Intention Estimation
  8. Backup Control Barrier Function Modification
  9. Human Interface
  10. Implementation
  11. Hardware
  12. Backup Controllers
  13. Results
  14. Conclusions and Future Work

Key Result

Theorem 1

ames2017control If $h$ is a CBF for system on $\mathcal{C}$, then any locally Lipschitz continuous controller ${\mathbf{k}: \mathbb{R}^n \to U}$ satisfying renders eq:clsystem1 safe with respect to $\mathcal{C}$.

Figures (5)

  • Figure 1: Visualization of the proposed safety-critical control framework with desired robot behavior. The red, green, and blue arrows represent different trajectories that rely on different backup controllers, with corresponding colors, to guarantee safety. Using driver intention tracking, the correct, green, backup controller is chosen among the red and blue controllers in order to guide the driver to their desired location. A supplementary video can be found here: https://youtu.be/41Jh1GD_9Ok
  • Figure 2: Cross entropy and training loss during a training episode. Accuracy is calculated by comparing the maximum reward ouput from the LSTM-DNN architecture and determining if it corresponds to the correct choice of backup controller for the respective point in the training set. We also use a validation (test) dataset to observe out-of-sample performance of the network during training. The network achieves 97% accuracy by the 30th epoch.
  • Figure 3: Our test-vehicle (US Army DEVCOM GVR-Bot robot) in the NASA Jet Propulsion Laboratory Mars Yard. $(yellow\ inset)$ Intel Realsense D457 Depth Cameras are coupled with a VN-100 IMU, $(light-blue\ inset)$ custom compute payload.
  • Figure 4: Robot trajectory which used all three backup controllers under our learned switching law. Subfigure (a) shows the robot trajectory, where direction is indicated by the arrows and the choice of backup controller is indicated by color. Subfigures (b), (c), and (d) show the segments of the robot trajectory that used $\mathbf{k_{b0}}, \mathbf{k_{b1}}, \mathbf{k_{b2}}$ respectively. In (b), (c), and (d) we also show the flows of the respective backup controllers at several robot positions. Notice that in each of (b) (c) (d), the backup controller flow always escapes the obstacle. However, notice that $\mathbf{k_{b0}}$ may not provide safe evasion from the obstacle in the locations in subfigure (c) as it does for the locations in (b). This implies that our system chose the correct backup controller depending on several factors, like driver intent and robot position.
  • Figure 5: Demonstration of system safety during switches between backup controllers. See that $h(x) \geq 0$ and the input constraints are satisfied (denoted by the gray dashed lines); therefore, our system maintains safety while better aligning with driver intention.

Theorems & Definitions (3)

  • Definition 1: Control barrier function ames2017control
  • Theorem 1
  • Theorem 2