Obstacle Affordance Detection using Perception Sensors in Robotics
by Paul van Houtum/
This literature survey will cover state-of-the-art work on object affordance detection methods, which will be introduced using a small review on 3D sensor data representation and object state detection methods. More specifically, the main contribution of this work lays in summarizing the
existing work on object affordance detection, by categorizing the methods by their used model learning types. Another key contribution of this work is, placing the roboticists’ perception of affordances in perspective, towards the psychologist Gibson’s original definition of affordances from a biological point of view. Finally, the use case of affordances in mobile manipulator robots will briefly be enlightened.
The published work on 3D sensor data representation methods, shows that choosing a well-performing method boils down to a compromise between accuracy and computational speed. These are generally key requirements for object state detection methods, that exist in various approaches. However, these methods tend to have a great difficulty differentiating
between permanently static objects and static objects that are able to be moved. In this context, affordances are introduced as a solution to this problem, by being able to visually detect whether an object is able to be displaced. The work found in this field, tends to approach affordances in another way compared to Gibson’s definition of affordances. Because, most of the work focuses on detecting affordances of objects without incorporating the robot capabilities, as required by Gibson’s definition. Finally, the incorporation of affordances in interactive navigation scenarios using mobile manipulator robots, is promoted as being promising.