Walking in Narrow Spaces: Safety-Critical Locomotion Control for Quadrupedal Robots with Duality-Based Optimization

Qiayuan Liao, Zhongyu Li, Akshay Thirugnanam, Jun Zeng, Koushil Sreenath

TL;DR

This paper introduces a novel safety-critical locomotion control framework for quadrupedal robots that enables them to navigate extremely tight spaces. By combining duality-based optimization with discrete-time control barrier functions (DCBFs), the robot can precisely model its own shape and obstacles as polytopes, leading to significantly less conservative obstacle avoidance than previous methods. This allows for faster, more agile navigation in cluttered environments, demonstrated on a real quadrupedal robot.

What Problem Does This Paper Address?

Quadrupedal robots, with their inherent maneuverability, are ideal for traversing complex and rough terrains. However, controlling them safely, especially in confined or cluttered environments, remains a significant challenge. Existing approaches often rely on simplified, conservative approximations of the robot and obstacles (e.g., spheres or circles) to ensure safety. While this can work in open spaces, it leads to overly cautious behavior, potential deadlocks, and an inability to navigate very narrow passages. The core problem this paper addresses is enabling quadrupedal robots to perform safety-critical locomotion in tight spaces without sacrificing speed or agility, by improving the precision of obstacle avoidance.

Methods

The paper proposes a Nonlinear Model Predictive Control (NMPC) framework that integrates exponential Discrete-time Control Barrier Functions (DCBFs) with duality-based optimization for obstacle avoidance.

  • Polytopic Representation: Instead of spherical approximations, the robot and obstacles are represented as polytopes (geometric shapes defined by linear inequalities). This allows for a much more accurate representation of their actual shapes.
  • Duality-based Distance Computation: Calculating the minimum distance between polytopes is non-smooth. The paper leverages the principle of duality from optimization to reformulate the primal distance minimization problem into a dual problem that can be solved efficiently as a Quadratic Program (QP). This dual formulation is key to handling polytope-based obstacle avoidance.
  • Exponential DCBFs: To integrate safety constraints into the NMPC, the authors use exponential DCBFs. These constraints ensure that a safety function (related to the distance to obstacles) decays exponentially over time, guaranteeing that the robot remains within a safe set.
  • NMPC Integration: The duality-based polytope distance constraints are incorporated into the NMPC as exponential DCBF constraints. This allows the robot to plan its trajectory while simultaneously considering its dynamics, the desired path, and precise obstacle avoidance.
  • Whole-Body Control (WBC): The optimized states and inputs from the NMPC are then fed into a Whole-Body Controller (WBC) to compute the actual motor torques for the robot.

Datasets

The paper does not explicitly mention using pre-existing datasets for training. Instead, it focuses on real-world experiments and simulations to validate its proposed framework. The "datasets" used are essentially the environments and scenarios created for testing, including:

  • Simulated environments with single and random obstacles.
  • Real-world experiments involving navigating through straight, L-shaped, V-shaped, and randomly cluttered narrow corridors.

Key Insights and Ideas

  • Polytopes for Precision: Representing the robot and obstacles as polytopes, rather than spheres, is crucial for enabling tight maneuvers. This finer geometric representation directly translates to less conservative avoidance and the ability to navigate much narrower spaces.
  • Duality as an Enabler: The duality formulation is a clever mathematical tool that transforms a computationally challenging non-smooth distance calculation between polytopes into a solvable QP, making polytope-based avoidance practical within an MPC framework.
  • CBFs for Smoothness and Safety: Exponential DCBFs provide a robust way to enforce safety constraints within the NMPC, leading to smoother trajectories and preventing collisions even in dynamic situations. The paper shows that adding CBFs has a negligible impact on computation time.
  • End-to-End System: The integration of perception (pointcloud to polytopes), planning (NMPC with DCBFs), and control (WBC) creates a complete autonomy stack capable of real-world deployment.

Key Results

  • Tighter Navigation: The proposed method allows the robot to navigate significantly tighter spaces compared to methods using spherical approximations. For example, in a narrow corridor scenario, the robot successfully navigated where a spherical approximation-based method failed (8.3% failure rate vs. 100% failure rate).
  • Faster Traversal: The robot achieved faster travel times in benchmark tests (e.g., 8.4s vs. 13.5s on Map 1) due to less conservative path planning.
  • Smoother Trajectories: The use of CBFs resulted in smoother robot trajectories, particularly when approaching obstacles, avoiding abrupt changes in heading.
  • Negligible Computational Overhead: The addition of exponential DCBF constraints increased computation time by only about 1.55 ms on average, making it suitable for real-time online deployment.
  • Real-World Validation: The framework was successfully deployed on a real quadrupedal robot (Unitree A1), demonstrating its ability to autonomously and safely navigate through various challenging narrow environments.

Limitations and Future Work

  • 2D Navigation Focus: The current work primarily focuses on 2D navigation. Future work aims to extend this to 3D obstacle avoidance with polytopes.
  • Robustness to Control Errors: The authors acknowledge that control errors can occur and suggest future work to incorporate robustness, potentially by adding DCBF Duality constraints into the WBC.
  • Global Planner Limitations: In some cases, failures were attributed to an infeasible global path that did not fully consider the robot's configuration. Improving the global planner's awareness of robot dynamics could further enhance performance.

Summary updated

Abstract

This paper presents a safety-critical locomotion control framework for quadrupedal robots. Our goal is to enable quadrupedal robots to safely navigate in cluttered environments. To tackle this, we introduce exponential Discrete Control Barrier Functions (exponential DCBFs) with duality-based obstacle avoidance constraints into a Nonlinear Model Predictive Control (NMPC) with Whole-Body Control (WBC) framework for quadrupedal locomotion control. This enables us to use polytopes to describe the shapes of the robot and obstacles for collision avoidance while doing locomotion control of quadrupedal robots. Compared to most prior work, especially using CBFs, that utilize spherical and conservative approximation for obstacle avoidance, this work demonstrates a quadrupedal robot autonomously and safely navigating through very tight spaces in the real world. (Our open-source code is available at github.com/HybridRobotics/quadruped_nmpc_dcbf_duality, and the video is available at youtu.be/p1gSQjwXm1Q.)
BibTeX
@inproceedings{liao2023walking,
  author = {Qiayuan Liao and Zhongyu Li and Akshay Thirugnanam and Jun Zeng and Koushil Sreenath},
  title = {{Walking in Narrow Spaces: Safety-Critical Locomotion Control for Quadrupedal Robots with Duality-Based Optimization}},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems},
  year = {2023},
  doi = {10.1109/iros55552.2023.10341896},
}
Figure