Safe Adaptive Switching among Dynamical Movement Primitives: Application to 3D Limit-Cycle Walkers

Sushant Veer, Ioannis Poulakakis

Abstract: Complex motions for robots are frequently generated by switching among a collection of individual movement primitives. We use this approach to formulate robot motion plans as sequences of primitives to be executed one after the other. When dealing with dynamical movement primitives, besides accomplishing the high-level objective, planners must also reason about the effect of the plan's execution on the safety of the platform. This task becomes more daunting in the presence of disturbances, such as external forces. To alleviate this issue, we present a framework that builds on rigorous control-theoretic tools to generate safely-executable motion plans for externally excited robotic systems. Our framework is illustrated on a 3D limit-cycle gait bipedal robot that adapts its walking pattern to persistent external forcing.

Introduction

In modern robotics, complex behaviors are often constructed by stitching together simpler "Dynamical Movement Primitives" (DMPs). These pre-programmed skills—such as walking straight, turning, or sidestepping—allow for modular motion planning. However, a significant challenge arises during the transition between these primitives. While a robot may be stable within a single behavior, the act of switching can introduce instabilities, particularly in unpredictable environments.

Real-world robots must contend with persistent external disturbances, such as wind or physical interaction with humans. If a robot switches behaviors too rapidly or at an inappropriate state under these conditions, it risks exiting its "safe zone" (basin of attraction) and falling. This research introduces a framework to provide rigorous safety guarantees for switching without requiring the robot to have prior knowledge of the exact external forces it will encounter.

Methodology: A Hierarchical Control Framework

The proposed approach treats the robot as a switched system with multiple equilibria. The framework is organized into a hierarchical structure that separates high-level decision-making from low-level execution.

The hierarchical system architecture: a high-level Supervisor manages switching logic based on state feedback, while the Low-level Control executes the selected movement primitive.

Technical Pillars

  1. Lyapunov Stability & SoS Programming: The framework uses Lyapunov functions to define "safe sets" for each movement primitive. To handle the complexity of 3D robotic systems, Sum-of-Squares (SoS) programming is employed to estimate the Basins of Attraction (BoA)—the regions where stability is guaranteed.
  2. Average Dwell-Time (ADT): The core logic relies on a calculated "buffer time" known as average dwell-time. This mathematical lower bound dictates how long a robot must remain in one primitive before switching to the next, ensuring that the energy dissipated by stable dynamics outweighs any instability introduced by the switch.
  3. Disturbance Agnosticism: A key theoretical contribution is the proof that if switching logic is safe for an unperturbed system, it remains safe under small, persistent disturbances. The supervisor does not need to model the disturbance; it only needs to understand the robot's inherent stability properties.
  4. Hybrid Zero Dynamics (HZD): To make these proofs computationally feasible for a 17-dimensional 3D walker, HZD is used to project the dynamics onto a 2-dimensional "restricted" space, significantly reducing dimensionality without losing essential stability information.

Experimental Validation: 3D Bipedal Walking

The framework was validated using a high-fidelity simulation of a 3D limit-cycle bipedal robot with nine degrees of freedom. The robot was tested in a "collaborative transport" scenario where it had to follow a leader's path based solely on interaction forces (pushes and pulls).

3D simulation of the bipedal walker successfully following a curved path by adaptively switching walking primitives in response to external forces.

Key Results

  • Stride-to-Stride Switching: For the 3D walker, the calculated average dwell-time allowed the robot to safely switch its walking primitive (e.g., from straight to turning) every single step.
  • Robust Path Following: The biped successfully followed a leader moving at 0.65 m/s along a curved path. By treating physical interaction forces as signals to decide the next primitive, the robot remained within its safe state-space regions throughout the task.
  • Stability under Pressure: The framework maintained stability even under persistent external forcing without needing a precise model of those forces, demonstrating significant robustness compared to non-adaptive methods.

Conclusion and Future Work

This research demonstrates that formal safety proofs, often limited to simple models, can be scaled to complex 3D robots. By balancing the "energy" of switching with a calculated dwell-time, robots can safely adapt to human collaboration and environmental disturbances.

Future work will focus on relaxing the "small disturbance" assumption to handle larger collisions and integrating the framework with obstacle-avoidance logic in more cluttered environments. Additionally, the authors aim to explore active disturbance estimation to further reduce the conservativeness of the current dwell-time bounds.

Summary updated

Abstract

Complex motions for robots are frequently generated by switching among a collection of individual movement primitives. We use this approach to formulate robot motion plans as sequences of primitives to be executed one after the other. When dealing with dynamical movement primitives, besides accomplishing the high-level objective, planners must also reason about the effect of the plan's execution on the safety of the platform. This task becomes more daunting in the presence of disturbances, such as external forces. To alleviate this issue, we present a framework that builds on rigorous control-theoretic tools to generate safely-executable motion plans for externally excited robotic systems. Our framework is illustrated on a 3D limit-cycle gait bipedal robot that adapts its walking pattern to persistent external forcing.
BibTeX
@inproceedings{veer2019safe,
  author = {Sushant Veer and Ioannis Poulakakis},
  title = {{Safe Adaptive Switching among Dynamical Movement Primitives: Application to 3D Limit-Cycle Walkers}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2019},
  url = {https://openreview.net/forum?id=OhdXf050TZ},
}
Figure