The goal of robotics is to design machines to assist or substitute humans in difficult tasks. To enable machines to properly interface with a physical process it is necessary to have information regarding the process state. Improving the ability of machines to measure a target's process state is a crucial step to improve the ability and autonomy of robots. To reduce uncertainty when inferring a process state, measurements from multiple sensing agents can be combined. One interesting question then is how to position the different sensing agents in order to maximize information yield. This is further complicated when considering sensing agents that have heterogeneous movement and sensing capabilities. The problem of reducing uncertainty about a physical process by designing optimal sensing trajectories for a team of mobile robots is known as the Active Information Acquisition (AIA) Problem.
This is an active area of research. In order to obtain novel results, we aim to combine two previous works. The first work on Cooperative target tracking for heterogeneous robots[2] solves the AIA problem using a Kalman filter for state estimation and a Nonlinear Model Predictive Controller (NMPC) to obtain sensing trajectories.
The second work on Motor and Perception Constrained NMPC for Torque-Controlled Generic Aerial Vehicles[1] uses an NMPC which considers real actuation limits to directly produce torque inputs for generic aerial platforms.
I will try to combine these works by extending the NMPC formulation in the work by Baljinder Singh Bal to include the motion model and actuator constraints of the second work by Martin Jacquet. To my knowledge using this NMPC formulation, to design control inputs for a team of heterogeneous robots in order to minimize uncertainty about multiple targets' process states is a novel approach.
The tentative research question is: Considering the Active Information Acquisition problem, how to design control inputs for a team of heterogeneous robots in order to minimize uncertainty about multiple targets' process states, what is the effect of using explicit nonlinear robot motion models and actuator constraints in a Nonlinear Model Predictive Controller?
[1] Martin Jacquet and Antonio Franchi. \Motor and Perception Constrained NMPC for Torque-controlled Generic Aerial Vehicles". In: IEEE Robotics and Automation Letters 6.2 (2020), pp. 1-1. ISSN: 23773766. DOI: 10.1109/lra.2020.3045654.
[2] Baljinder Singh Bal. Cooperative target tracking for heterogeneous robots. 2019.