COMP 765: Intelligent Robotics, Winter 2019
COMP 765 is a research seminar on Intelligent Robotics: the intersection of rocket science and the latest machine learning on-board systems that can sense the world and act upon it. The class will begin with lectures on definitional problems and algorithms in robotics. We will then transition to mixed student-lead and instructor-lead discussions of recent developments in research and in practice. The emphasis is on algorithms, probabilistic reasoning, learning to improve behaviors using data, and decision making under uncertainty, as opposed to electromechanical systems design.
- Jan 9, 2019 Please fill out the Welcome Survey which will be used to tailor the seminar to interests of the group this year.
- Jan 7, 2019 Welcome to the course!
OverviewWe will broadly cover the following areas:
- Sensing and Estimation: robots (like humans) sense their surroundings through limited, imprecise measurements. What algorithms are needed to give robots the ability to reason accurately about the world?
- Control and Planning: how can we compute the state-(in)dependent commands that can bring a robotic system from its current state to a desired state, when models of the robot and its environment are available?
- Decision Making Under Uncertainty: how can the same problems be solved when our model knowledge is missing or imprecise? In particular, what motion strategies best complement our probabilistic estimators, for example by gathering useful information or acting conservatively?
- Learning for Robotics: how can we use data to learn models for a robot's continuous time series data in case these are not known in advance?
- Heirarchical Decision Making: how can tasks with longer time-scale or for larger target systems be factored productively into multiple levels of reasoning?
- Human-robot Interaction: how can we make programming robots easy? Can robots learn from the way humans perform and describe tasks? Can we achieve super-human performance? Can robots teach humans how to best perform or reason about tasks?
The class will be a mix between lecture-style teaching and student-lead discussion. In each unit, I will lead things off by presenting standard material given in textbooks, covering the theory and analysis with basic examples. We will then transition to the state-of-the-art in the area, with the end of the unit featuring presentations of important recent papers following the PROPONENT, OPPONENT, REPRODUCER format to be described in class.
NOTE: Following the seminar course style, the material is taylored to the interests of the class each year, and there are numerous opportunities for student-lead discussion. For now this is a rough initial schedule to allow course selection, to be refined after the Welcome Survey and discussion during the first few classes.
Motivation and course sylabus
1 - Introduction
||PR Chapter 1.|
Kinematics and Dynamics
|2 - Spatial Representation (pdf) (pptx)||PR Chapter 5.|
Estimation under uncertainty
Sensing uncertainty, Bayesian filters
|3 - Estimation Intro||PR Chapters 2.|
Derivation and analysis, implementation concerns, EKF, sigma-points
|4 - Particle and Kalman Filters||PR Chapter 3|
Importance sampling, analysis, efficient resampling methods
|4 - Particle and Kalman Filters||PR Chapter 4|
Modern Localization and Mapping
Rao-Blackwellized Particle Filtering, large-scale sparse graph optimizers, mapping an entire city.
5 - Mapping
||PR Section 3|
|Jan 28 to Feb 13||
Control and Planning
Control theory, ODE background, PID, MDPs, dynamic programming, LQR, trajectory optimization, stability, controllability, POMDPs. Discrete planning: search and DP, symbolic planning. Continuous planning: sampling-methods, geometric approaches.
|PLAN text, selected chapters|
|Feb 18 to March 13||Learning for Robotics
Building blocks: Linear regression, Gaussian Processes, Bayesian Deep Nets. Learning to control: Model-based RL, End-to-end Deep RL. Active learning, black-box optimization applied to probabilistic search.
|GPforML text, selected chapters|
|March 18 to April 3||
Interacting with Humans
Imitation learning including LfD, IRL and AfP. HRI for home robots, self-driving and robots in the workplace. Safety assessment of intelligent robots. Collaborative and Cloud Robotics.
|Dec 5||Project Demonstrations and Wrap-Up|
- 3 assignments worth 10% each = 30%
- 3 in-class presentations worth 10% each = 30%
- The first as a PROPONENT, who presents the main content of the paper and gives its positive aspects
- The second as an OPPONENT, who focuses on the weaknesses or limitations of the paper, perhaps drawing on aspects where the related work is stronger
- The third as a REPRODUCER, who has tried the authors code or data (or re-implemented aspects of the paper) and reports on this experience
- 1 final project written report worth 40%
Recommended, but optional, textbooks
- Probabilistic Robotics, by Thrun, Fox, and Burgard (PR in the outline)
- Planning Algorithms, by Lavalle (PLAN in outline)
- Gaussian Processes for Machine Learning, by Rasmussen and Williams (GPforML in the outline)
- Computational Principles of Mobile Robotics, 2nd edition, by Dudek and Jenkin
- Robotics, Vision, and Control, by Corke
- Introduction to Autonomous Mobile Robots, by Siegwart, Nourbakhsh, Scaramuzza