Using intelligent robots to autonomously explore the environment is an attractive topic for both civil and military applications, such as search and rescue, surveillance and reconnaissance. Online planning for robot trajectory is a key component for successful environment exploration. We specifically focus on target search and tracking using autonomous robots.
Due to the lack of knowledge on the target state a prior, it is necessary for robots to move in an informative way to effectively gathers information of the target. We propose a model predictive control (MPC)-based path planning approach for ground mobile robots. By using the informative-theoretic objective function, the robot can plan an effective way for motion planning. Target estimation is conducted using Bayesian filters (e.g. particle filters and Kalman filters). The proposed method can also handle the intermittent measurements caused by the limited sensing domain.
With robots stepping into daily life of human beings, various applications for improving human life qualify has been envisioned. We are interested in enabling robots to work as companions of humans, walking with a target human while carrying items for the human. Especially, we have developed a trajectory planning algorithm for robots to autonomously follow a target human, using onboard cameras.
Due to the uncertainty in human's motion, we proposed a Parallel Interacting Multiple Model-Unscented Kalman Filter (PIMM-UKF) approch for human motion estimation and prediction. Based on the predicted human states, an MPC path planner is developed to produce safe and comfortable following trajectory of the robot.
Human-robot Interaction (HRI) has been an increasingly popular research area due to the boom in personal and industrial robots. We are especially interested in enabling robots to collaborate as peers with the human.
On one hand, we utilize Bayesian inference approach from Psychology community to help robots identify human's intention and then provide assistance accordingly. On the other hand, we model how human anticipates the collaborative robot's plan based on the partial actions that the robot has made. These two aspects close the interaction loop between human and robot and can be useful for developing algorithms to improve human-robot collaboration.
Humans usually team up with robots in a hierarchical structure, i.e., humans work as operators or supervisors to give commands to robots. However, there is an increasing necessity for humans to work with robots as peer partners, such as the collaborative assembly and search and rescue.
We focused on the collaborative search of a target using a team of humans and robots. Robots constantly infer and predict human plan and provide assistance accordingly.
Distributed filtering using multiple robots has many important applications such as environment monitoring, target localization, and field state estimation. Information sharing composes a key component for distributed algorithms.
We proposed a measurement dissemination-based distributed Bayesian filter for non-centralized estimation of target position under fixed or time-variant communication topology. Such method has been shown to generate consistent estimation and fast convergence performance.