EECS 495 - Machine Learning and Artificial Intelligence for Robotics
COURSE DESCRIPTION: A coverage of artificial intelligence, machine learning and statistical estimation topics that are especially relevant for robot operation and robotics research. The focus is on robotics-relevant aspects of ML and AI that are not covered in depth in EECS 348 or EECS 349. Course evaluation will be largely project-based.
COURSE COORDINATOR: Prof. Brenna Argall
PREREQUISITES: Graduate-level standing (or permission of instructor) for the maths, some programming experience (in Matlab okay).
1. Crash course in robotics: sensors and sensing, effectors and actuators
2. Probability basics
II. State estimation and uncertainty filters
1. Bayes filters
2. Gaussian filters : Kalman, Information...
3. Particle filters
III. Machine Learning
1. Bayesian Learning : Bayes rule, Bayes classifier, MAP, MLE, EM, Mixtures of Gaussians...
2. Linear classifiers : perceptron, winnow...
3. Experts style:voting, bandits...
4. Programming: Linear, Quadratic, Convex
5. Genetic Algorithms
6. InstancebasedLearning : nearest neighbors, regression (linear, locally weighted, kernel)...
7. Reinforcement Learning : Bellman, Qlearning,TDlearning, actorcritic...
IV. Artificial Intelligence
2. Informed : Greedy, A*, D*, heuristic functions...
3. Local/optimizing : gradient descent, hill climbing, simulated annealing...
2. Behavior based robotics : reactive, subsumption architecture, hierarchical control...
V. Special topics
- Week 0 : Introduction
- Week 1 : State estimation and uncertainty filters
- Week 2 : ML: Bayesian Learning, Linear Classifiers, Expertsstyle
- Week 3 : ML: Programming, Genetic Algorithms
- Week 4 : ML: InstancebasedLearning
- Week 5 : ML: Reinforcement Learning
- Week 6 : AI: Planning
- Week 7 : AI: Search, BehaviorbasedRobotics
- Week 8 : Project presentations
- Week 9 : Project presentations, Special topics