[Paper Review] Team CERBERUS Wins the DARPA Subterranean Challenge: Technical Overview and Lessons Learned

Marco Tranzatto, Mihir Dharmadhikari+34 more|arXiv (Cornell University)|Jan 1, 2022

Robotics and Automated Systems15 citations

TL;DR

This paper presents CERBERUS, a system-of-systems integrating legged and flying robots with multi-modal perception, resilient autonomy, and unified path planning to win the 2021 DARPA Subterranean Challenge. The system achieved robust exploration, accurate artifact detection, and high-fidelity mapping in GPS-denied, sensor-degraded underground environments using a human-supervised, single-command interface.

ABSTRACT

This article presents the CERBERUS robotic system-of-systems, which won the DARPA Subterranean Challenge Final Event in 2021. The Subterranean Challenge was organized by DARPA with the vision to facilitate the novel technologies necessary to reliably explore diverse underground environments despite the grueling challenges they present for robotic autonomy. Due to their geometric complexity, degraded perceptual conditions combined with lack of GPS support, austere navigation conditions, and denied communications, subterranean settings render autonomous operations particularly demanding. In response to this challenge, we developed the CERBERUS system which exploits the synergy of legged and flying robots, coupled with robust control especially for overcoming perilous terrain, multi-modal and multi-robot perception for localization and mapping in conditions of sensor degradation, and resilient autonomy through unified exploration path planning and local motion planning that reflects robot-specific limitations. Based on its ability to explore diverse underground environments and its high-level command and control by a single human supervisor, CERBERUS demonstrated efficient exploration, reliable detection of objects of interest, and accurate mapping. In this article, we report results from both the preliminary runs and the final Prize Round of the DARPA Subterranean Challenge, and discuss highlights and challenges faced, alongside lessons learned for the benefit of the community.

Motivation & Objective

• To develop a resilient, autonomous robotic system-of-systems capable of exploring complex, GPS-denied subterranean environments.
• To enable high-level human supervision with a single operator managing multiple robots in dynamic, sensor-degraded conditions.
• To achieve reliable detection and localization of 40 artifacts within 60 minutes in diverse underground scenarios.
• To overcome challenges in perception, navigation, communication, and mobility through integrated multi-modal sensing and unified planning.
• To advance robotic autonomy for real-world applications in search and rescue, disaster response, and underground exploration.

Proposed method

• Employed a hybrid fleet of legged (ANYmal C) and flying robots (RMF-Owl, Alpha Aerial Scout, Kolibri) for complementary mobility and perception.
• Integrated multi-modal perception using visual-inertial, LiDAR, thermal, and radio frequency sensors to maintain localization and mapping under sensor degradation.
• Implemented a unified exploration and local motion planning framework that accounts for robot-specific capabilities and limitations.
• Deployed self-deployable communication-extender modules from legged robots to extend network range in communication-denied environments.
• Utilized a tethered roving platform as a mobile communication hub to enhance network resilience.
• Developed robust control and safety mechanisms in behavior trees, including foothold safety thresholds and recovery behaviors based on field testing.

Experimental results

Research questions

• RQ1 How can a heterogeneous team of legged and flying robots achieve resilient autonomy in GPS-denied, sensor-degraded subterranean environments?
• RQ2 What multi-modal perception and mapping strategies enable reliable localization and object detection under degraded visual conditions such as fog, dust, and darkness?
• RQ3 How can unified path planning and control systems account for robot-specific mobility constraints and terrain challenges?
• RQ4 What role do self-deployed communication extenders play in maintaining network connectivity during exploration?
• RQ5 What lessons from field testing can improve the robustness of autonomy and perception systems in real-world subterranean deployments?

Key findings

• CERBERUS won the DARPA Subterranean Challenge Final Event, successfully locating all 40 artifacts within the 60-minute time limit.
• The system achieved accurate artifact detection and localization within 5 meters of ground truth for all correctly reported artifacts.
• Field testing revealed critical sensor degradation issues (e.g., self-similarity, fog, smoke), which guided the development of multi-modal localization and mapping frameworks.
• The introduction of batched motion cost prediction and foothold safety thresholds significantly improved path planning safety and reduced rare outliers in real-world execution.
• Open-sourcing key components like GBPlanner2, Maplab, and Voxblox, along with simulation models and a full dataset from the Prize Round, enables broader community reuse and advancement.
• The tethered roving platform and communication-extender deployment proved essential for maintaining network connectivity in large-scale, multi-level underground environments.