Learning Joint Space Reference Manifold for Reliable Physical Assistance

TL;DR

The paper tackles reliable physical STS assistance with the Talos humanoid despite large, unpredictable human forces by learning a 1D spatial reference manifold that maps force input to joint configurations. It uses a functional representation (Bernstein polynomials) to parameterize a low-dimensional trajectory manifold and defines a robust cost that accounts for ZMP constraints across a range of forces, enabling offline learning of a stable spatial reference. Validation includes simulations and real-robot experiments showing reduced joint-space discontinuities and improved resilience to human variability, though limitations remain in direct ZMP control and online adaptation. Future work points to online MPC on the manifold, direct ZMP feedback, and higher-dimensional manifolds with joint optimization of the coordination structure.

Abstract

This paper presents a study on the use of the Talos humanoid robot for performing assistive sit-to-stand or stand-to-sit tasks. In such tasks, the human exerts a large amount of force (100--200 N) within a very short time (2--8 s), posing significant challenges in terms of human unpredictability and robot stability control. To address these challenges, we propose an approach for finding a spatial reference for the robot, which allows the robot to move according to the force exerted by the human and control its stability during the task. Specifically, we focus on the problem of finding a 1D manifold for the robot, while assuming a simple controller to guide its movement on this manifold. To achieve this, we use a functional representation to parameterize the manifold and solve an optimization problem that takes into account the robot's stability and the unpredictability of human behavior. We demonstrate the effectiveness of our approach through simulations and experiments with the Talos robot, showing robustness and adaptability.

Figures (5)

• Figure 1: Experimental setup. (a) Force and human articulation position data collection during human-human interaction (only force data is utilized in this paper), (b) Robot's dynamic model, (c) Simulation setup where the human is modeled as an external force, and (d) human-robot interaction scenario.
• Figure 2: Exploring the impact of diverse steps on varying force magnitudes.
• Figure 3: The force that the user applied at one of his/her hands when interacting with another user.
• Figure 4: Examining the performance of the two systems operating under different force magnitudes $M \ [N]$ (columns), and its rising time $h \ [s]$ (rows). The force is applied for $2h$ seconds. Green and gray blocks represent successful and failure motions, respectively.
• Figure 5: The forces users applied at each robot's hand (the first row), and their corresponding ZMP change (the second row). Different columns correspond to different users.