果冻传媒

My name is Hao Liang Chen and my research topic focuses on making mobile robots more explainable and robust in semi-open environments, i.e. structured environments in which we know what kind of disturbances we can expect but where the exact realization of these disturbances is unknown in advance. This research takes place in the Control Systems Technology research group of the Department of Mechanical Engineering.

NEW COMBINATION OF KNOWLEGDES Optimum actuator forces

An example of a semi-open environment would be a robot driving through a corridor, in which the structured environment is represented by static walls (and doors) and the disturbances by (vague) notions like 鈥榙oors being opened鈥�, 鈥榩eople walking around鈥� or 鈥榖oxes being put on the ground鈥�. My approach is to give the robot some common-sense knowledge to allow it to deduce its desired behavior, e.g. 鈥渨alls are solid, hence I cannot go through them鈥�. A step further is to combine pieces of knowledge in order to deduce the context of a situation. For example, pieces of knowledge like 鈥渕y name is being called鈥�, 鈥渢he person calling me is looking at me鈥� and 鈥渢he person is carrying something鈥� may be combined to deduce that a person is calling the robot to help them with something.

The challenge, meanwhile, is determining what kind of knowledge to give the robot. One could, for example, give the knowledge 鈥測ou should go to location A鈥�, but the pitfall is that this cannot really be reused for other applications nor is it explainable. Why should the robot go to location A? A better approach is to decompose the knowledge into two pieces, like 鈥淚 need to do action X鈥� and 鈥渓ocation A allows me to perform action X.鈥� Via these pieces of knowledge, the robot can deduce that it should go to location A.

MOVING BEYOND 'IF-ELSE'
 

As robots become more present in our daily lives, they need to be able to deal with all the possible variations in our environments. Giving robots some common-sense knowledge allows them to better integrate into our society and facilitates situations in which robots may cooperate with humans. This research is therefore interesting for people who want to deploy mobile robots in semi-open environments in which they also want the robot to be able to explain/understand why it did something in a certain way. One shouldn鈥檛 need to pre-program every action sequence the robot could take in an 鈥榠f-else鈥� fashion 鈥� due to the many variations possible in the real world, we also hypothesize that it鈥檚 impossible to pre-program everything. 

THE BUILDING BLOCKS OF SELF-AWARENESS
 

In this project, I work with two other PhD students and one postdoc student. One of the PhD students, Bob Hendrikx, focuses more on the localization and world modeling aspects, i.e. giving the robot the tools for some sense of awareness. I will then provide the reasoning aspect to let the robot itself determine how to use these tools. Another PhD student, Margot Neggers, focuses on human-robot interactions, providing me with knowledge of how the robot should behave in the presence of humans. The reasoning aspect of the robot should then use this knowledge to adjust behavior. The postdoc student, Liang Li, focuses more on the learning aspect of this project, so we can collaborate in the future on how the robot should learn new knowledge. As for now, I鈥檓 busy implementing and testing my approach in a use-case.