[Paper Review] NeBula: Quest for Robotic Autonomy in Challenging Environments; TEAM CoSTAR at the DARPA Subterranean Challenge

Ali Agha, Kyohei Otsu+70 more|arXiv (Cornell University)|Mar 21, 2021

Robotics and Automated SystemsEngineering105 citations

TL;DR

NeBula is an uncertainty-aware, belief-space autonomy framework enabling resilient, modular robotic exploration for DARPA Subterranean Challenge, demonstrated on heterogeneous robots across Tunnel and Urban competitions.

ABSTRACT

This paper presents and discusses algorithms, hardware, and software architecture developed by the TEAM CoSTAR (Collaborative SubTerranean Autonomous Robots), competing in the DARPA Subterranean Challenge. Specifically, it presents the techniques utilized within the Tunnel (2019) and Urban (2020) competitions, where CoSTAR achieved 2nd and 1st place, respectively. We also discuss CoSTAR's demonstrations in Martian-analog surface and subsurface (lava tubes) exploration. The paper introduces our autonomy solution, referred to as NeBula (Networked Belief-aware Perceptual Autonomy). NeBula is an uncertainty-aware framework that aims at enabling resilient and modular autonomy solutions by performing reasoning and decision making in the belief space (space of probability distributions over the robot and world states). We discuss various components of the NeBula framework, including: (i) geometric and semantic environment mapping; (ii) a multi-modal positioning system; (iii) traversability analysis and local planning; (iv) global motion planning and exploration behavior; (i) risk-aware mission planning; (vi) networking and decentralized reasoning; and (vii) learning-enabled adaptation. We discuss the performance of NeBula on several robot types (e.g. wheeled, legged, flying), in various environments. We discuss the specific results and lessons learned from fielding this solution in the challenging courses of the DARPA Subterranean Challenge competition.

Motivation & Objective

• Motivate and address autonomous exploration in unknown, challenging subterranean environments.
• Present the NeBula framework and its uncertainty-aware, modular design for resilient decision-making.
• Showcase performance and lessons learned from Team CoSTAR in DARPA Subterranean Phase I (Tunnel) and Phase II (Urban).
• Demonstrate applicability across heterogeneous robot platforms and environments (ground, aerial, hybrid).
• Discuss components from perception to mission planning and communication that enable large-scale, decentralized autonomy.

Proposed method

• Present NeBula as an uncertainty-aware belief-space autonomy framework that plans over joint distributions of robot pose, environment, and mission states.
• Describe the modular architecture linking perception, planning, world belief management, communications, and operations.
• Introduce SMAP (simultaneous mapping and planning), SLAP (simultaneous localization and planning), and SPLAM concepts to enforce perception-aware decision-making.
• Detail multi-sensor state estimation with HeRO (Heterogeneous and Resilient Odometry) and sensor fusion pipelines across LiDAR, IMU, vision, thermal, and other modalities.
• Explain the team’s ConOps for heterogeneous robot deployment, mesh networking, dynamic task allocation, and data sharing under intermittent communications.

Experimental results

Research questions

• RQ1 How can autonomy be made resilient in perceptually degraded and communication-denied subterranean environments?
• RQ2 What are the benefits of planning over belief spaces (joint inference and decision-making) for robust exploration and artifact detection?
• RQ3 How does NeBula perform across heterogeneous robots and environments (tunnel, urban, planetary analogs) in the SubT Challenge?
• RQ4 What are the system-level lessons and trade-offs when deploying large-scale, networked autonomous teams in challenging terrains?

Key findings

• CoSTAR achieved 2nd place in the Tunnel competition and 1st place in the Urban competition.
• NeBula was demonstrated on multiple robot types, including wheeled, legged, flying, and hybrid platforms, across terrestrial and analog planetary environments.
• The architecture emphasizes resiliency through joint perception-planning, uncertainty handling, and modularity for scalable, networked autonomy.
• Field demonstrations and lessons learned highlight the importance of resilient state estimation, large-scale mapping, semantic understanding, and bandwidth-aware communication.
• The paper discusses practical insights on heterogenous system integration, simulator-based development, and transitioning from supervised to autonomous operation.