Flexible perception-motivated action plans to improve system autonomy

• Grant number: 12/19898-4
• Support Opportunities: Scholarships abroad - Research
• Start date: April 15, 2013
• End date: January 29, 2014
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Carlos Henrique Quartucci Forster
• Grantee: Carlos Henrique Quartucci Forster
• Host Investigator: Brian Charles Williams
• Host Institution: Divisão de Ciência da Computação (IEC). Instituto Tecnológico de Aeronáutica (ITA). Ministério da Defesa (Brasil). São José dos Campos , SP, Brazil
• Institution abroad: Massachusetts Institute of Technology (MIT), United States

Abstract

Active perception is a means to improve the quality of information obtained from the environment through the execution of actions motivated by perception improving, as a result, the outcomes of an automatic decision. Simpler autonomous systems can be constructed to work in structured environments, by sensing and responding reactively and adapting to unexpected events that can be easily detected. In order for an autonomous system to cope with the uncertainty, incomplete knowledge and dynamicity of richer and less predictable environments, perception has to be improved. Perception-motivated plans are able to diminish uncertainty of the environment by guiding the perception process. We propose an approach to active perception modeling the system with three essential components: a sensorial subsystem with low-level interpretation algorithms, a model-based planner and executive that is able to run goal-oriented adaptable plans and a conceptual unit that comprehends the generation or update of goals and constraints aimed at improved perception. Those perception goals are inspired in related problems such as active learning, active probing and active estimation. Our experiments employ object recognition as the perception modality, addressing the problems of partial observability due to occlusion, projection and resolution. They happen in a physical setting simulating an industrial manufacturing scenario with a rich and dynamic environment and robots are expected to move occluding obstacles, bring objects closer to the camera, observe from different positions or ask for help. (AU)