|
doi:10.3850/978-981-08-7304-2_1618
Efficient Learning of a Robot for Autonomous Exploration Refined using Behavior-based Hybrid Architecture
Dip N. Ray1, Amit Mondal1, Sumit Mukhopadhyay2 and Somajyoti Majumder1
1S. R. Lab, Central Mechanical Engineering Research Institute (CSIR), Durgapur, W. B.-713209, India.
2Department of Mechanical Engineering, National Institute of Technology, Durgapur, W. B.-713209, India.
ABSTRACT
The paradigm of robotics can be classified in two categories: Classical robotics and Behavior based robotics. Behavior based robotics uses Sense → Act methodology rather than Sense →Plan → Act methodology of classical robotics. Behavior based robots provide good response to stimuli. Here a behavior based architecture has been used to simulate the exploration in an maze. This refined data has been used in real robot learning. Learning refers to systematic design and development of algorithms to evolve behaviours. Reinforcement learning is a reward- punishment based learning with constant interaction for robots. The generated data from simulation has been used in a real robot for learning to explore a similar maze. The results for explorations using Behavior based architecture and Q-learning are nearly equal.
Keywords: Behavior based robotics, Reinforcement learning, Q-learning, Hybrid architecture.
Back to TOC
FULL TEXT(PDF)
|