Self-reflective perceptual adaptation for Spot to recognize and follow a human teammate

Lifelong autonomy introduces extra difficulties to robot perception, including sudden and long-term changes of the environment (e.g., lighting changes) and sensor failures. We propose a new human-inspired self-reflective perceptual adaptation method that is able to adjust robot perception according to both the environment context and the robot's sensing capability. Our method jointly performs feature learning, sensor fusion, and perception calibration under a unified regularized optimization framework. [IJRR Paper | 12/08/2021]

Enhancing consistent ground maneuverability by robot adaptation to complex off-road terrains

During ground navigation, a robot's planned maneuvering behaviors cannot always be accurately executed due to setbacks such as reduced tire pressure. This inconsistency negatively affects the robot's maneuverability and can cause slower traversal time or errors in localization. We propose a method for consistent behavior generation that learns offset behaviors in a self-supervised fashion to compensate for this inconsistency without requiring the explicit modeling of various setbacks, while adapting to a variety of complex off-road terrains. [CoRL Paper | 11/23/2021]

Bayesian deep graph matching for CoID in collaborative perception

In collaborative perception, correspondence identification (CoID) aims to identify the same objects to ensure consistent object references by a team of agents in their own fields of view. We propose an uncertainty-aware deep graph matching method that formulates CoID as a deep graph matching problem under the Bayesian framework to explicitly quantify and reduce correspondence uncertainty and perceptual non-covisibility during learning. Our method is evaluated in the applications of human-robot collaborative assembly and connected driving. [RSS Paper | 07/12/2021]

Watch a Spot robot explore an old mine

We brought our robotic dog Spot from Boston Dynamics to the Edgar Mine for testing its out-of-box capabilities. This initial testing was featured by CNET: "The ground is rocky and uneven. Old, rusted rails that used to carry loads of precious metals run the length of the path. Most wheeled robots would have trouble navigating this uneven surface, but it's not a problem for Spot." This Spot robot will be mainly used for robot learning and adaptation research in the HCR lab. [CNET | 10/09/2020]

Leading multi-agent teams to multiple goals while maintaining communication

We introduce the problem of multiple robot teammates tasked with leading a multi-agent team to multiple goal positions while maintaining communication. We define utilities of making progress towards goals, maintaining communications with followers, and maintaining communications with fellow leaders. We propose a novel regularized optimization formulation that balances these utilities and utilizes structured sparsity inducing norms to focus the leaders' attention on specific goals and followers over time. [RSS Paper | 07/12/2020]

Correspondence identification under uncertainty in collaborative perception

Correspondence identification is essential for multi-robot collaborative perception, which allows a group of robots to consistently refer to the same objects in their own fields of view. It is challenging problem caused by non-covisible objects that cannot be observed by all robots and the uncertainty in robot perception. We propose a principled approach of regularized graph matching that addresses perception uncertainty and non-covisibility in a unified mathematical framework to identify the correspondence. [RSS Paper | 07/12/2020]

AR-based communication for synchronized and time-dominant human-robot teaming

We presented a design framework that uses a Unity-ROS simulation for developing augmented reality (AR) strategies to improve communication in synchronized, time-dominant human-robot teaming. This work has the potential to impact any domain in which humans conduct synchronized multi-domain operations alongside autonomous robots in austere environments, including search and rescue, disaster response, and homeland defense. [05/18/2020]

Engineers develop robots that are able to do dangerous jobs

"Every day it seems, robots acquire new capabilities. The robotics field continues to evolve rapidly. Now, robots are starting to go underground to explore mines, caves and tunnels that can be pretty dangerous to humans." We are developing a team of autonomous and collaborative ground and aerial robots for underground and field reconnaissance. This research was reported by CGTN AMERICA. [07/22/2018]