Surprise-Based Learning for Autonomous Systems
Abstract
Dealing with unexpected situations is a key challenge faced by autonomous robots. This paper describes a promising solution for this challenge called "Surprise-Based Learning" (SBL). in which a learning robot engages in a life-long cyclic learning process consisting of "prediction, action, observation, analysis (of surprise) and adaptation". In particular, the robot always predicts the consequences of its actions, detects the surprises whenever there is a significant discrepancy between the prediction and the observed reality, analyzes the surprises for its causes and uses the critical knowledge extracted from analysis to adapt itself to the unexpected situations. SBL was successfully demonstrated on a physical modular robot which learned to navigate to desired goal-scenes with no prior knowledge about the environment, its sensors or the consequences of its actions. The scalability of SBL in the number of sensors and actions was evaluated to be reasonable for a physical robot, while adaptation to unexpected situations such as hardware failure and goal transfer was very successful.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 28, 2009
- Accession Number
- ADA585802
Entities
People
- Nadeesha Ranasinghe
- Wei-min Shen
Organizations
- University of Southern California