Dynamic Walking on Slippery Surfaces : Demonstrating Stable Bipedal Gaits with Planned Ground Slippage

Wen-Loong Ma, Yizhar Or, Aaron D. AmesICRA 2019 · DOI

TL;DR

This paper presents the first experimental demonstration of stable bipedal robot walking on slippery surfaces by explicitly planning for controlled ground slippage. By extending hybrid system dynamics to include stick-slip transitions and using nonlinear programming for gait optimization, they enable robots to walk robustly on challenging terrains where traditional no-slip assumptions fail.

What Problem Does This Paper Address?

Bipedal robot locomotion has made significant strides, but a fundamental assumption in most research and development is that the robot's foot maintains a no-slip contact with the ground. This assumption holds true on high-friction surfaces but breaks down dramatically on slippery terrains, leading to instability and failure. This paper tackles the challenge of enabling robust bipedal walking on such low-friction environments, which are common in natural settings. The authors aim to move beyond simply avoiding slippage or recovering from it as a disturbance, and instead, to embrace and strategically utilize controlled slippage as a means to achieve stable locomotion.

Methods

The core of this work lies in extending the theoretical framework for bipedal locomotion to explicitly account for stick-slip dynamics. This involves:

  • Hybrid System Modeling: The robot's walking motion is modeled as a hybrid system, incorporating multiple continuous-time domains (sticky walking and slippery walking) and discrete events (impacts, stick-slip transitions). Coulomb's friction law is used to model the transitions between sticking and slipping.
  • Trajectory Optimization: A nonlinear programming (NLP) approach, specifically using the FROST toolbox, is employed to generate optimal walking gaits. This optimization considers the complex hybrid dynamics, including physical limitations and the feasibility of controlled slippage.
  • Feedback Control: Virtual constraint-based controllers are used for real-time stabilization. While the optimization is based on feedback linearization, the experimental implementation uses a PD controller for improved robustness.

Datasets

Not discussed in the paper. The experiments were conducted on a custom treadmill setup with varying levels of lubricant to simulate slippery surfaces.

Key Insights and Ideas

  • Slippage as a Feature, Not a Bug: The paper demonstrates that planned, controlled slippage can be a viable strategy for stable locomotion on slippery surfaces, rather than just an undesirable disturbance.
  • Energy Efficiency: Slipping gaits, when optimized, can potentially be more energy-efficient than their non-slipping counterparts, even on non-slippery surfaces, by reducing impact forces and body kinetic energy changes.
  • Robustness through Adaptability: The AMBER-3M robot, equipped with the optimized slipping gait, showed remarkable robustness and adaptability to varying degrees of slipperiness, even walking stably on a non-lubricated surface.
  • Bridging Theory and Experiment: This work successfully bridges the gap between theoretical advancements in modeling and optimizing for slippery locomotion and their practical experimental realization on a physical robot.

Key Results

  • The AMBER-3M robot was able to achieve stable, dynamic walking on a lubricated treadmill by employing a pre-planned slipping gait.
  • A conventional "no-slip" gait, which works on a rough treadmill, failed on the slippery surface.
  • The optimized slippery gait demonstrated robustness to varying levels of lubrication and even worked on a non-lubricated surface.
  • Simulations showed that a slippery gait could be up to 140% more energy-efficient than a sticky gait on a sticky surface, with reduced impact velocities and body kinetic energy changes.
  • Experimental measurements of mechanical cost of transport (MCOT+) showed that while the robot could walk on slippery surfaces, the experimental efficiency was significantly lower than simulated values, suggesting further research is needed to bridge this gap.

Limitations and Future Work

  • Experimental MCOT+ Discrepancy: The experimental energy efficiency was significantly lower than simulated values, and the differences between various slippery conditions were less pronounced than expected. The authors attribute this to the dominance of the robot's nominal energy usage and environmental uncertainties.
  • Ordered Sequence of Transitions: The optimization formulation pre-specified a particular sequence of transitions (e.g., stick to slip). Future work could explore optimizing these sequences more dynamically.
  • Comprehensive Energy Study: A more comprehensive study on energy consumption for bipedal robots walking on slippery surfaces is needed, isolating the effects of slippage from nominal energy usage and environmental factors.
  • Exploring Different Slippage Patterns: Further research could investigate different ordered sequences of domains and a more comprehensive study on energy consumption for bipedal robots walking on slippery surfaces.

Summary updated

BibTeX
@inproceedings{ma2019dynamic,
  author = {Wen-Loong Ma and Yizhar Or and Aaron D. Ames},
  title = {{Dynamic Walking on Slippery Surfaces : Demonstrating Stable Bipedal Gaits with Planned Ground Slippage}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2019},
  doi = {10.1109/icra.2019.8793761},
}
Figure