Information Theory, Inference & Learning Algorithms
Information Theory, Inference & Learning Algorithms
A fast learning algorithm for deep belief nets
Neural Computation
Towards hierarchical blackboard mapping on a whiskered robot
Robotics and Autonomous Systems
| Hi-index | 0.00 |
Behaviours in autonomous agents --- animals and robots --- can be categorised according to whether they follow predetermined stimulus-response rules or make decisions based on explicit models of the world. Probability theory shows that optimal utility actions can in general be made only by considering all possible future states of world models. What is the relationship between rule-following and optimal utility modelling agents? We consider a particular case of an active mapping agent using two strategies: a simple wall-following rule and full Bayesian utility maximisation via entropy-based exploration. We show that for a class of environments generated by Ising models which include parameters modelling typical robotic mazes, the rule-following strategy tends to approximate optimal action selection but requires far less computation.