Autonomous mapping and target identification exercises in Mammoth Cave.
Credit: Image from JPL’s Costar Team
The rapid advancements in autonomy and robotics (including sensing and mobility) has progressively reduced the gap of achieving fully autonomous exploration of harsh and hard-to-reach environments.
One example of technology advancements in that direction is the DARPA Subterranean Challenge (SubT), a robotic competition (September 2019 through September 2021) that pushed several technology barriers towards autonomous robotic exploration of kilometer-long underground environments including natural cave networks, tunnel systems, and urban underground infrastructure. Subterranean environments in particular pose significant challenges for unmanned operations due to limited communication, situational awareness, power resources, and visibility (e.g. fog, dust).
In SubT, robot teams are required to map, navigate, and search massive environments for particular objects of interest (e.g. mannequin survivors, backpacks, cell phones, helmets, etc), called artifacts, within a limited amount of time (e.g. 30 minutes to 1 hour).
NASA JPL’s Team CoSTAR has developed new advance technologies and a novel autonomy framework, Nebula, for addressing several challenges critical for enabling autonomous exploration of large and unknown underground voids with large number of heterogeneous robots, including autonomous robot coordination and exploration, risk-aware navigation, communication and networking, artifact detection, and advance localization and mapping. For more information on these technologies (including publication, videos), please visit our CoSTAR team website
The JPL AI Group supported this effort by developing new technologies for mission autonomy, situational awareness and automated operations.
One particular rule in the SubT competition stands out and pushes participants to rely on autonomy: the entire team of robots must be controlled by a single operator. This requirement in particular brings interesting operability challenges: while the massive underground environment and limited time of exploration bring the need for a large team of robots, the deployment and operations of a large number of robots can go beyond the cognitive capacity of a single human operator. For example, an operator would have a very hard time to efficiently control 2-3 robots at the same time in these environments, let alone controlling 8-15 robots.
In order to support the single operation to manage this challenging multi-robot coordination endeavor, we developed an autonomous assistant for human-in-the-loop multi-robot operations, called Copilot MIKE. The assistant has been developed to help the operator to 1) rapidly setup and deploy multiple to maximize exploration time (the less time we spend setting up the more time we will have to map and detect objects of interest), and 2) command the robots during exploration with minimal intervention while maintaining a bearable workload and high situational awareness. This is a key set of capability in robotic exploration with limited time and resources.
This work leads to improved human-supervisory control for a multi-agent system reducing overhead from application switching, task planning, execution, and verification while increasing available exploration time with a human-autonomy teaming platform.
The DARPA SubT competition had its final event in September 2021. Efforts to evaluate and generalize application of Copilot MIKE algorithms have followed.
TeamSubset of CoSTAR team that participates in this effort:
Tiago Stegun Vaquero (Mission Planning Team Lead), Jet Propulsion Laboratory, California Institute of Technology
Ali Agha (CoSTAR team lead), Jet Propulsion Laboratory, California Institute of Technology
Giovanni Beltrame, Polytechnique Montreal
Gustavo Correa, Jet Propulsion Laboratory, California Institute of Technology
Marcel Kaufmann, Jet Propulsion Laboratory, California Institute of Technology
Michael Milano, Jet Propulsion Laboratory, California Institute of Technology
Kyohei Otsu, Jet Propulsion Laboratory, California Institute of Technology
Maira Saboia, Jet Propulsion Laboratory, California Institute of Technology
Ryan Stonebraker, Jet Propulsion Laboratory, California Institute of Technology
Robert Trybula, Jet Propulsion Laboratory, California Institute of Technology
JPL Artificial Intelligence Working Group (JPL AIWG)