Taking the First Step Toward Autonomous Quadruped Robots: The Quadruped Robot Challenge at ICRA 2023 in London [Competitions]

Adam Jacoff, Jeongmin Jeon, Oliver Huke, Dimitrios Kanoulas, Seehoon Ha, Donghyun Kim, Hyungpil MoonIEEE Robotics Autom. Mag 2023 · DOI

Introduction

The inaugural Quadruped Robot Challenge (QRC) at ICRA 2023 established a new benchmark for legged locomotion, testing robots against "real-world" obstacles like sinking foam and slippery rails. While remote control remains the industry standard for field operations, the competition highlighted the significant technical hurdles—and the massive scoring incentives—driving the community toward fully autonomous navigation in complex environments.

Participants and their quadruped robots at the ICRA 2023 competition.

While quadruped robots like those from Boston Dynamics or Unitree have become commercially available, bridging the gap between a controlled lab environment and a chaotic field site remains a massive challenge. Researchers often solve specific sub-problems (like gait or mapping) in isolation. The QRC was designed to force an integration of these technologies, addressing the lack of standardized, reproducible testing grounds for legged robots. Specifically, it tackles the "autonomy gap": the difficulty of combining motion control, localization, and mapping to navigate terrains that violate basic assumptions, such as the ground being a rigid, non-slippery plane.

Methods and Arena Design

The organizers developed a modular, reconfigurable arena consisting of five distinct sections, each targeting a specific robotic weakness. To encourage autonomy, the scoring system applied a 4x multiplier to autonomous runs compared to remote-controlled ones. Teams were given 10-minute windows to complete as many trials as possible, with a 2-minute penalty for manual resets.

Arena overview and diagram showing the five challenge sections (A-E).

The five sections included:

  • Section A (Ramp Terrains): 15° slopes to test traction and balance.
  • Section B (Soft Floor): 10-cm thick foam designed to simulate sand or water, where feet sink below the perceived ground plane.
  • Section C (Pallet Steps with Pipes): Elevation changes with rolling pipe edges to trip up wheeled or tracked systems.
  • Section D (Slippery Floor with K-Rails): Plywood floors with diagonal rails to test foot placement and slip recovery.
  • Section E (Crate Terrain): Positive and negative obstacles (holes) that challenge a robot's perception and "blind" stepping capabilities.

Key Insights and Results

Despite the 4x scoring incentive, only two out of the six participating teams actually attempted autonomous runs. This underscores that robust autonomy in non-trivial environments remains a frontier for the research community.

The soft foam floor (Section B) proved to be a significant hurdle. Robots that rely on rigid-contact models for odometry often fail when the "ground" moves or absorbs the foot, requiring more sophisticated foot-slip compensation. Additionally, the popular Unitree Go1 robots faced specific physical challenges, such as rear legs getting caught on K-rail barriers and difficulty perceiving negative obstacles (holes).

Team KAIST emerged as the Overall Champion and Best in Autonomy. They successfully navigated the terrain using map-based planning and inertial visual odometry with specialized foot-slip compensation. Team MIT secured first place in the Remote Control category.

The arena layout was deemed so effective for testing operator proficiency that it is being transported to the U.K. RACE (Remote Applications in Challenging Environments) facility for permanent use in training robot operators for hazardous environments.

Limitations and Future Work

The authors acknowledge that current success in autonomy relied heavily on pre-mapped environments and the structural help of side walls. Future iterations of the QRC (Stage 2) will move away from "just walking" toward task-oriented challenges, such as active search for inspection tasks, mobile manipulation with robotic arms, and multi-robot coordination in shared spaces.

Summary updated

BibTeX
@article{jacoff2023taking,
  author = {Adam Jacoff and Jeongmin Jeon and Oliver Huke and Dimitrios Kanoulas and Seehoon Ha and Donghyun Kim and Hyungpil Moon},
  title = {{Taking the First Step Toward Autonomous Quadruped Robots: The Quadruped Robot Challenge at ICRA 2023 in London [Competitions]}},
  journal = {IEEE Robotics Autom. Mag.},
  year = {2023},
  doi = {10.1109/mra.2023.3293296},
}
Figure