Networked decentralized multi-robot systems has a wide variety of large-scale applications such as search and rescue, exploration of unknown environments and environmental sampling.
We propose a novel decentralized and behavior-based approach for a large group of robots moving in unknown environments with obstacles. We prove that (a) the robots will never lose connectivity while coordinating or collide with one another or obstacles during moving, and (b) the motion strategy is robust to insertion and failure of individual robots.
A. Li, W. Luo, S. Nagavalli and K. Sycara, "Decentralized Coordinated Motion for a Large Team of Robots Preserving Connectivity and Avoiding Collisions" IEEE International Conference on Robotics and Automation 2017 (ICRA2017) (Submitted)
In many scenarios involving human interaction with a remote swarm, the human-swarm communication channel is often extremely bandwidth constrained and may have high latency. For good human-swarm interaction, a summary representation can be generated by selecting a subset of robots, known as the information leaders, whose own states gives an approximation of the entire swarm.
We propose fully distributed asynchronous algorithms for information leader selection under state uncertainty that only rely on inter-robot local communication. We provide bounded optimality analysis and proof of convergence for the algorithms.
A. Li, W. Luo, S. Nagavalli, N. Chakraborty and K. Sycara, "Handling State Uncertainty in Distributed Information Leader Selection for Robotic Swarms" In Proceedings of the IEEE Conference on System, Man, and Cybernetics 2016 (SMC16), October, 2016
One of the most important complex Cyber-Physical Systems (CPS) is the physiology of the human body. In this project, we are working on designing experiments and methodologies to model Human-CPS system under a medical fluid management task.
We conducted human experiments, dicussed results, and proposed a compuational model based on Recurrent Neural Network (RNN) with Long Short Term Memory (LSTM) architecture to model human action in this human-CPS system with high fidelity.
|Structure from motion (SFM) problem is a broadly studied problem in computer vision. The objective for structure from motion is to recover the 3D structure of a target using multiple frames of images from different prospectives. We solve both extrinsic and intrinsic parameters of camera by solving a batch optimization problem that minimizes the reprojection error of SIFT features between frames.|
|In many human robot interaction senarios, non-verbal commands can be more intuitive and introduce less ambiguity than verbal commands. One example is assigning navigation goals to mobile robots, in which case a single pointing gesture could be very handful. In the Robotics Lab at Univerisity of Alberta, we built an interface that helps human to interact with a Segway robot using pointing gestures.|
|Vision system is essential to home environment service robots. One of the most important task is for the robot to recognize detect and recognize the object that the robot is supposed to serve, for example, bottled water, Coca-Cola, Sprite, book, etc. We trained a classifier with Speeded Up Robust Features (SURF) to detect and recognize objects placed on a table. The system is implemented under Robot Operating System (ROS).|