Teams will develop and demonstrate physical systems to compete in live competitions on physical, representative subterranean courses, and focus on advancing and evaluating novel physical solutions in realistic field environments.
Teams will develop software and algorithms using virtual models of systems, environments, and terrain to compete in simulation-based events, and explore larger-scale runs in simulated environments that explore significantly expanded scenario sizes and durations.
Inspiration has been mainly from inefficiencies in the current systems and how leaders in the robotics field have gone about solving similar problem with inefficiencies.
For the current “problem in hand” degraded communication would play a key role in order to report the identified artifacts effectively, we plan to solve this by having an optimal network of multiple robots able to communicate back to base as soon as an artifact is identified.
For artifact detection & identification, multiple sensor modalities will be utilized considering the unfavorable conditions for vision-based detection. Two object identification networks will be run. A network that utilizes LIDAR point clouds to detect and classify artifacts will be run in parallel with a camera vision-based detection which fuses data to derive 3D location of the artifacts. The two networks’ outputs will be compared against each other for a final detection solution, and adopted accordingly.
For guidance navigation and control we will follow one or more map, behavior or communications strategies. Behaviors such as mapping, localization, path planning, obstacle avoidance, following, goal seeking, formation that are derived from sensors form the basis for actuators. Different filters for controls and estimation would be applied depending on the error of any of the sensors and their performance.
Michigan Technological University/Michigan Tech Research Institute (MTU/MTRI)
Our team views the virtual track of the SubT Challenge as a problem in multi-agent coordination in highly resource-constrained settings. Resources in this case include agent lifespan, sensing ability, communications connectivity, among others. Our solution is inspired by the need to optimize the joint capabilities of the team as well as the utilization of their resources. We are also leveraging the mathematical strengths of our team to develop principled, generalizable, and novel solution strategies.
University of Nevada, Reno
ETH Zurich, Switzerland
University of California, Berkeley
Sierra Nevada Corporation
Team CERBERUS draws conceptual inspiration from the mythological cerberus, the three headed dog protecting the underworld. From a robotics standpoint we are inspired from the potential of combining two very different modalities of locomotion, namely walking and flying, while simultaneously addressing in a unified manner the perception and navigation challenges.Website
Scientific Systems Company, Inc.
Our system is built with the idea of allowing low-Size, Weight, and Power(SWaP) systems to take a key role in the exploration and navigation tasks. We designed our algorithms with low-SWaP systems in mind, which we believe separates our approach from many others.Website
Jet Propulsion Laboratory
California Institute of Technology
Massachusetts Institute of Technology
KAIST, South Korea
Our proposed solution blends several key components that have been in development at JPL, Caltech, and MIT for cave exploration and other applications. (1) For mobility, our hybrid aerial and ground vehicle, the “rollocopter”, arose from the need of having an energy-efficient platform that can also overcome obstacles in extreme environments. (2) Autonomous coordination of a team of robots has been of increasing interest for space and terrestrial missions, and we are leveraging our R&D in networked multi-agent autonomy (swarms) to maintain communications connectivity and explore efficiently and robustly. (3) As part of a multi-robot solution, we will solve distributed simultaneous localization and mapping (SLAM), enabling the system to fuse individual sensor measurements into a joint solution of what the environment looks like and where all the robots are located within it. (4) Because JPL often seeks solutions for locations without GPS (like Mars), we have been developing a magnetic quasi-static (MQS) system to aid in medium-range localization by placing and reading magnetic fields. (5) Communications in underground environments can be especially challenging; JPL has a research program specifically looking at networks and communications waveforms for cave exploration.Website
Subterranean systems frequently have vertical passages. People traverse those using ropes. Ground robots can’t currently that approach. So, a flying robots are used that can deal both the rubble on the floor and vertical passages. UAVs of course have their own set of disadvantages such as limited cargo capacity and short battery life. Our approach is to mitigate those weaknesses and take advantage of the UAV’s strengths.
Commonwealth Scientific and Industrial Research Organisation, Australia
Georgia Institute of Technology
Our approach to the challenge relies on a blend of our team's core capabilities: mobility in extreme terrains, localisation & mapping in GPS-denied environments, and reliable navigation in previously unknown environments. By combining CSIRO Data61, Emesent, Georgia Tech and other Australian partner universities and robotic companies' expertise, we aspire for a seamless cooperation between robotic agents in complex environments.Website
Czech Technological University, Czech Republic
Université Laval, Canada
Jersey Media Network Corp.
We want to come out of this project with a reliable solution for deploying underground robots either for exploration or for search and rescue operations. We’re combining our advanced experience in deep learning with the specifics of this challenge – precise localization and mapping of an unknown environment without a GPS and with limited communications, and swarm robotics algorithms for optimal graph exploration with multiple agents.Website
Carnegie Mellon University
Oregon State University
Our approach is inspired by observing many fragile robot demos and seeing that our solutions for exploration often don't match the mobility and size constraints for an actual deployment. Therefore, our approach pursues the themes of resilience and modularity. Resilience will allow our robots to even perform in cases where the nominal approach fails, and modularity will us to rapidly reconfigure platforms for relevant environments.Website
Sophisticated Engineering UG
Using two eyes together with the recognition of the own movement a human being can get an intuitive impression of the surrounding area. Rebuilding this method in a robot by using only cameras (no laser measurement) and movement sensors was our inspiration.
To follow this approach we implemented a new configuration for the challenge using two cameras on the robots. Based on the camera and sensor data each robot navigates through the tunnel autonomously and builds up a map of the tunnel system. A team of multiple robots are sent out to search the tunnel system simultaneously to save time. The robots use an individual communication layer on top of the system communication layer to exchange information between the robots when they are within the reception area of another robot.
University of Colorado, Boulder
University of Colorado, Denver
Scientific Systems Company, Inc.
We aim to combine local autonomy and multi-agent distributed goal satisfaction through the synchronization of distilled navigation and goal information over a robust custom mesh network designed to support large numbers of nodes under highly transient connectivity. Our solution is inspired by search and rescue teams and aims to mirror the human ability to distill information down to the minimum required for efficient coordination and communication and to synchronize that information in as efficient a manner as possible to numerous agents under difficult conditions.Website
National Chiao Tung University
Our approaches are deeply inspired by our recent work in the Duckietown & AI Driving Olympics 2018-2019, and the Maritime RobotX Competition 2018. Our long-term flight, collision-safe aerial blimp robot Duckiefloat is adapted from the Duckebot in the Duckietown platform, the miniaturized testbed to develop autonomy education and research of a fleet of robots. Duckiefloat mimics the lane following in Duckietown for tunnel following, and altitude controls from our underwater robot developed in the RobotX Competition. The Anchorball launcher on the UGV Husky is adapted from the one on our WAM-V (wave adaptive modular vehicle) in RobotX. We continue developing a revised Duckietown, including multiple robots with similar setups to Duckiefloat, artifacts, and tunnel-like environments, as a testbed for DARPA SubT Challenge.
University of Pennsylvania
In considering the variety of mobility challenges that we would face in the subterranean environment we felt that it would be best to use a team of heterogeneous agents both legged and flying to take advantage of the advantages of both approaches. Legged systems can provide enhanced mobility over rough terrain and extended mission durations while aerial vehicles are well suited for exploring complex 3D environments. Our approach to coordination is inspired in part by economic systems which also involve multiple agents that need to coordinate their actions in the face of uncertainty.
Robotika International, Czech Republic and United States
Robotika.cz, Czech Republic
Czech University of Life Science, Czech Republic
Centre for Field Robotics, Czech Republic
Cogito Team, Switzerland
eXtreme Programming: What is the simplest solution which could possibly work (score at least one point)? Now we think that it could be following the wall there and back again...Website
The system is designed for simplicity in the behavior of each robot; the goal being for the robots to go and explore without getting in each other's way or repeatedly exploring the same areas more than necessary. Searching for artifacts involves treating each observation as an unreliable estimate of an artifact's identity and location, which is refined further over time. The system is also designed to be versatile enough to be able to operate using different robot bases and sensor configurations, requiring only minimal changes to the system configuration.