Planning Brachistochrone Hip Trajectory for a Toe-Foot Bipedal Robot going Downstairs

Gaurav Bhardwaj, Utkarsh A. Mishra, N. Sukavanam, R. BalasubramanianarXiv 2020 · arXiv

Introduction

Walking downstairs represents one of the most challenging locomotive tasks for bipedal robots. Unlike walking on level ground, descending stairs requires the management of significant gravitational potential energy and the maintenance of stability during high-impact transitions. Traditional trajectory planning methods for the hip—the robot's center of mass hub—typically rely on simple geometric paths such as circular arcs or linear "virtual slopes." While computationally efficient, these methods often result in high accelerations and jerk, which can destabilize the robot or cause excessive mechanical wear.

This research proposes a novel approach inspired by the principle of fastest descent: the brachistochrone curve. By modeling the hip's vertical motion after the path a particle would take to travel between two points in the shortest time under gravity, the researchers aim to achieve smoother, more natural, and more efficient robotic motion.

System Modeling and Methodology

The study utilizes a 9-link planar biped robot model. A key feature of this model is the "toe-foot" design, which includes joints for the hip, knee, ankle, sole, and toe. This configuration allows for more nuanced foot-ground interactions compared to rigid-foot models, enabling the robot to mimic the heel-to-toe transition seen in human gait.

The Brachistochrone Hip Trajectory

The core innovation lies in the hip trajectory planning. The vertical motion of the hip is formulated using the brachistochrone equation. This path is inherently optimized for gravity-assisted descent, allowing the robot to "fall" into the next step with minimal resistance. To maintain balance during this descent, the horizontal motion is governed by a Zero Moment Point (ZMP) based Center of Gravity (COG) trajectory, ensuring the robot's stability throughout the gait cycle.

Cycloidal Ankle Trajectory

To complement the hip's motion, the swing leg's ankle follows a cycloidal path. The mathematical properties of a cycloid ensure that the acceleration is zero at both the beginning and the end of the step. This is critical for "soft landings," as it minimizes the impact forces transmitted through the robot's frame when the foot contacts the stair tread.

Adaptive Planning and Neural Inverse Kinematics

The algorithm is designed to be adaptive, calculating the initial hip height based on the specific link lengths of the robot and the dimensions of the stairs. To solve the resulting Inverse Kinematics (IK) problem, the researchers employed an unsupervised Feed-Forward Artificial Neural Network (ANN). Unlike traditional IK solvers that require complex trigonometric derivations or large labeled datasets, this ANN is trained in real-time. Its loss function minimizes the Euclidean distance between the desired Cartesian coordinates and the position calculated via forward kinematics, allowing for efficient and flexible joint angle solutions.

Results and Comparative Analysis

The proposed brachistochrone method was validated through MATLAB simulations and compared against two standard trajectories: the Circular Arc and the Virtual Slope (linear) methods. The performance was measured based on the maximum absolute acceleration and jerk experienced by the robot's joints.

Metric Brachistochrone Circular Arc Virtual Slope
Max Abs. Acceleration 32 rad/s² 38 rad/s² 48 rad/s²
Max Abs. Jerk 2000 rad/s³ 2700 rad/s³ 4700 rad/s³

The results demonstrate that the brachistochrone path significantly reduces mechanical stress. It achieved a 15.8% reduction in acceleration compared to the circular arc and a 33.3% reduction compared to the virtual slope. Furthermore, the virtual slope method was found to be geometrically infeasible for certain stair configurations, failing to satisfy workspace constraints that the adaptive brachistochrone method handled with ease.

Discussion and Conclusion

By treating gravity as an ally rather than a force to be resisted, the brachistochrone trajectory planning offers a more fluid and human-like descent for bipedal robots. The combination of this physics-inspired path with cycloidal ankle trajectories and neural-network-based kinematics provides a robust framework for complex terrain navigation.

However, several avenues for future research remain. The current study is limited to 2D sagittal plane simulations and does not account for 3D dynamics or sensor noise. Additionally, while the kinematic results are superior, a full dynamic torque analysis is required to quantify the power savings for the robot's motors. Future work will also explore the role of torso dynamics, such as leaning, to further enhance stability during high-speed descents.

Summary updated

BibTeX
@misc{bhardwaj2020planning,
  author = {Gaurav Bhardwaj and Utkarsh A. Mishra and N. Sukavanam and R. Balasubramanian},
  title = {{Planning Brachistochrone Hip Trajectory for a Toe-Foot Bipedal Robot going Downstairs}},
  year = {2020},
  eprint = {2012.02301},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2012.02301},
}
Figure