Cognitive Robotics
Programming a cognitive robot: monitoring hidden state, flexible-time scheduling, planning, multi-agent and human collaboration, and risk-bounded programs.
Cognitive robotics applies autonomy to real-world systems, and this course builds up a cognitive robot’s program one reasoning capability at a time. It starts from the familiar paradigm of programming and adds monitoring of hidden and continuous state, scheduling tasks in flexible time, planning toward goal states, collaboration with other agents and with humans, and risk-bounded programming, until the robot reasons as an intelligent system.
What You Will Achieve
- Write model-based programs that monitor a robot’s hidden state
- Schedule tasks in flexible time and plan toward goal states
- Plan for multiple agents and for execution alongside humans
- Bound the risk of plans over time, choice and continuous state
Program Overview
Monitoring, Time and Goals
Programs that watch their own state, schedule and plan.
- Model-based programming and hidden state
- Flexible time and choice
- Planning as heuristic forward search
- Planning with temporal landmarks
Continuous State and Collaboration
Programs for continuous systems, teams of agents and people.
- Monitoring continuous state
- Continuous goal states
- Multi-agent planning
- Programs that execute with humans
Risk and the Grand Challenge
Risk-bounded programming and the group work that closes the term.
- Risk-bounded programs with flexible time and choice
- Risk-bounded programming on continuous state
- Student advanced lectures
- The grand challenge project
Course Completion
Job Outlook
Autonomy and planning methods are used in robotics, autonomous vehicles, spacecraft and aerospace systems, and AI research.
Good for: Students and engineers who want to program robots that plan, monitor themselves and work with people.
Keep in mind: Taught in 2016 in a field the instructor says reinvents itself every three years or so; the problem sets run in a course virtual machine, and the advanced lectures and grand challenge were group work.
Bottom line: A cognitive robot programmed one reasoning capability at a time, ending in risk-bounded planning.
Frequently Asked Questions
What do the problem sets cover?
Modeling exercises, existing autonomy tools and algorithm implementations, including conflict-directed A*, scheduling, hybrid state estimation, iterative risk allocation, incremental path planning and Monte Carlo tree search.
How do I run the assignments?
They use the Cognitive Robotics virtual machine, a 2 GB download in the Related Resources section, with instructions followed inline in IPython notebooks; Assignment 1 explains the installation.
What is the grand challenge?
A group project at the end of the term, driven by a research question in cognitive robotics, that involves modeling, implementing algorithms and debugging in a simulation environment.
Is there an interview with the instructor?
Yes. Professor Brian Williams explains what cognitive robotics is, why the course uses teams and a grand challenge, and how a lost Mars mission at NASA shaped his first grand challenge.



There are no reviews yet.