Fast scene analysis using vision and artificial intelligence for object prehension by an assistive robot. (August 2017)

• Record Type: Journal Article
• Authors: Bousquet-Jette, C.; Achiche, S.; Beaini, D.; Law-Kam Cio, Y.S.; Leblond-Ménard, C.; Raison, M.
• Is Part Of: Engineering applications of artificial intelligence. Volume 63(2017:Mar.)
• Page Start: 33
• Page End: 44
• Publication Date: 2017-08
• Subjects: OBJ 3D object file format -- PLY "Polygon File Format" -- STL "Standard Triangle Language" file format; Scene analysis -- Artficial vision -- Artificial intelligence -- Object grasping -- Assistive robotics; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2017.04.015
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: Robotic assistance for people affected by motor deficits is a fast growing field. In this context, two major challenges remain in terms of automated scene analysis and automated object prehension. More specifically, the most robust of current segmentation methods are still computationally intensive, preventing the automation of objects prehension from being fast enough to be considered acceptable as an everyday technical-aid. The objective of this study is to develop a fast scene analysis using vision and artificial intelligence for object prehension by an assistive robot. The solution developed in this paper aims at facilitating human-machine interaction by enabling users to easily communicate their needs to the technical aid. To achieve this, this paper proposes several novelties in three interconnected domains: scene segmentation, prehension and recognition of 3D objects. A novel technic, inspired by mechanical probing, is developed for scenes probing to detect objects. A simple, fast and effective decision tree is proposed for object prehension. Finally, the physical characteristics of the 3D objects are directly used in the neural network without using discriminants features descriptors. The results obtained in this paper have shown that scene analysis for robotic object prehension in cooperation with a user can be performed with effective promptness. Indeed, the system requires on average 0.6 s to analyze an object in a scene. With the JACO robotic assistance arm, the system can pick up a requested object in 15 s while moving at 50 mm/s, which may be greatly improved upon using a faster robot. The system performance averages 83% accuracy for object recognition and is able to use a decision tree to select a simple approach path for the robot end-effector towards a desired object. This system, in combination with an assistive robot, has great potential for providing users suffering from musculoskeletal disorders with improved autonomy and independence, and for encouraging sustained usage of this type of technical aids.