Planning and Representation Learning under Uncertainty for Motor Primitives and Robotics
Principal Supervisor: Dr. Vien Ngo
Second Supervisor: Prof. Neil Robertson
+ Project Description
The foremost objective of this research is to understand uncertainty factors and build learning and goal-directed behavior in intelligent systems working under uncertainty, in both artificial autonomous agent and natural robotic systems. We will tackle this goal by posing and answering two fundamental questions: i) how can appropriate internal representationbe learned by interaction under uncertainty with the environment efficiently? Especially in goal-directed robotic applications, how can motor primitives, behavior-based skills, and object relations be learned under uncertainty?; ii) how can behavior planning and motor control on any internal representation of uncertainty be realized? The second objective is potential applications of developed methods in robotics and motor control, which are manipulation under uncertainty (in which integrated systems of skill learning, planning, perception, state-estimation, and action in complex mobile manipulation domains are needed), and human-robot collaboration (in which human-intention, cooperation task-recognition, robust communication are important uncertainty factors).
We focus on robotic applications because robots are becoming more capable in autonomously solving complex daily tasks. Those tasks often require the robots to select actions over a long time horizon, e.g. over hours or days, under uncertainty about i) noisy or imperfect perception,
ii) properties and locations of objects, iii) partial information in human-robot interaction and iv) pervasive delayed dependencies. Specifically, we focus on problems where the environment can be described in terms symbolic or/and geometric states. The state might represent symbolic relations of objects, the degree-of-freedoms of environment, and kinematics of the possible motion of all objects. Actions can i) articulate objects or DoFs (degree of freedom) to change their symbolic or kinematic relations, or to discover new DoFs 1 and novel dependencies, ii) disambiguate uncertainty in interactive tasks with human.
+ How to Apply
Applicants should apply electronically through the Queen’s online application portal at: https://dap.qub.ac.uk/portal/
+ Contact Details
|Supervisor Name:||Dr. Vien Ngo|
+44 (0)28 9097 1824