Principles of Autonomy and Decision Making
AI and operations research for autonomy: search, constraints, planning, diagnosis, path planning, optimization, HMMs and MDPs.
How does an autonomous system, such as a spacecraft, a robot or a vehicle, decide what to do? This survey takes reasoning, learning and optimal decision-making methods from artificial intelligence and operations research, with a focus on principles, algorithms and their application: state space search, constraint programming, activity planning and execution, model-based diagnosis, path planning, linear and mixed-integer programming, reasoning under uncertainty with hidden Markov models, Markov decision processes and game theory.
What You Will Achieve
- Implement and analyze uninformed and informed search algorithms
- Model problems with constraints and propositional logic and solve them
- Plan activities and robot paths, including with sampling-based algorithms
- Make decisions under uncertainty with hidden Markov models and Markov decision processes
Program Overview
Search, Constraints and Planning
The foundations of state space search and activity planning.
- State space search, its complexity, soundness and completeness
- Constraint programming and satisfaction
- Activity planning with Graphplan
- Plan termination and execution
Model-Based Reasoning and Optimization
Logic, diagnosis, path planning and mathematical programming.
- Propositional logic and satisfiability
- Diagnosis, mode estimation and conflict-directed A*
- Informed search and sampling-based motion planning
- Linear and mixed-integer programming for vehicle routing
Uncertainty, Learning and Games
Inferring state and acting well when the world is uncertain.
- Hidden Markov models and the Baum-Welch algorithm
- Markov decision processes and policy iteration
- Sequential and differential games
Course Completion
Job Outlook
These methods are used in robotics, autonomous vehicles and spacecraft, operations research, logistics and planning software, and AI engineering.
Good for: Programmers who want a rigorous grounding in the classic algorithms behind autonomous decision making.
Keep in mind: Taught in 2010; the programming assignments use Java starter code, and only selected final exam problems come with solutions.
Bottom line: A broad, algorithm-first tour of how machines reason, plan and decide.
Frequently Asked Questions
What do the programming assignments involve?
Implementing search algorithms, backtrack search with forward checking, constraint optimization and A*, with incomplete Java code to finish, plus linear programming with lpsolve.
What does the planning project ask for?
Activity planning with SATPlan and path planning with rapidly-exploring random trees, using example code, PDDL files for a lunar logistics mission and environment files.
How does it relate to 16.413?
The graduate subject 16.413 met with this course, with more advanced programming and written assignments, including a Java tutorial.



There are no reviews yet.