recruitment
PhD studentship in “Flexible robotic control via co-operation between an operator and an AI based control system”
We’re currently advertising a PhD studentship in “Flexible robotic control via co-operation between an operator and an AI based control system”.
The project is funded for up to 4 years and is an industrial collaboration with DSTL (Defence Science and Technology Laboratories). Applicants must be EU citizens. The project would suit an Engineering / Computer Science graduate with an interest in Robotics and Human Factors or a Psychology graduate who is also a strong programmer with experience in robotics.
The student will be paid a full stipend for up to four years.
The main aim of this project is to design, implement and test new semi-autonomous control programs to enable a robot to perform tasks in collaboration with a remote human supervisor. The amount of control devolved to the robot will vary according to the task and may change dynamically throughout task execution.
This robot will fall between the two current extremes of fully autonomous robots performing highly constrained tasks, and robots with relatively little intelligence behaving under direct supervision by a human tele-operator.
Psychologic theories and methodologies will be used to measure and predict the cognitive requirements for both humans and robots involved in such collaborations. The aim is to compare the overhead needed to team humans and AI components and to identify ways of supplying mixed initiative control that improve overall task performance.
This project will evaluate the utility of various types of of adaptive, variable autonomy in a range of tasks, including taking account of cognitive demands on human supervisors. Currently deployed robots are generally of one of two types. Either they are entirely autonomous (computer controlled) or they are tele-operated by a human operator. Tele-operation allows an operator to control a robot from a safe distance, but as a result the operator’s situational awareness is impaired and they can have difficulty in controlling a vehicle, because they do not have the same situational awareness as if they were standing near the robot - communication and sensor limits often give them very narrow and low resolution fields of view. Fully autonomous robots on the other hand have made great strides in recent years, both in control of motion and manipulation. It is still difficult, however, to move seamlessly between autonomy of a robot control system and tele-operation intervention by a human supervisor, even in cases where local decision making by the robot using a wide range of sensors would exceed human tele-operative performance. Of a particular interest is the need for the operator to understand what the robot knows and what it is capable of as well as what it is likely to do. This cognitive overhead on the part of the operator can make using a more autonomous system harder than directly supervised remote control.
This project aims to understand which situations allow an Artificial Intelligence based decision making system to usefully control a robot in partnership with an operator.
Objectives include:
(a) Determine which situations are best tackled by autonomous AI or human
operator supervision and how to recognise these situations.
(b) Model the cognitive load on the operator in understanding the current system state and communicating new goal states to the robot.
(c) Design and build interfaces which make use of these models to improve overall performance.
(d) Explore a graded transition between variable levels of autonomy and human supervision.
Interested candidates should contact Rustam Stolkin (R.Stolkin@cs.bham.ac.uk) or Jeremy Baxter (J.Baxter@cs.bham.ac.uk).