CS 333 Safe and Interactive Robotics

CS 333: Safe and Interactive Robotics

Fall 2017-2018, Class: Tue, Thu 3:00-4:20pm, McMurtry 360

Description:

Once confined to the manufacturing floor, robots are quickly entering the public space at multiple levels: drones, surgical robots, service robots, and self-driving cars are becoming tangible technologies impacting the human experience. Our goal in this class is to learn about and design algorithms that enable robots to reason about their actions, interact with one another, the humans, and the environment they live in, as well as plan safe strategies that humans can trust and rely on.

This is a project-based graduate course that studies algorithms in formal methods, control theory, and robotics, which can improve the state-of-the-art human-robot systems. We focus on designing new algorithms for enhancing safe and interactive autonomy.

Format:

The course is a combination of lecture and reading sessions. The lectures discuss the fundamentals of motion planning, formal methods applied in robotics, learning from demonstration, intent inference, and shared control required for modeling and design of safe and interactive autonomy for human-robot systems. During the reading sessions, students present and discuss recent contributions in this area. Throughout the semester, each student works on a related research project that they present at the end of the semester.

Prerequisites:

There are no official prerequisites, but an introductory course in artificial intelligence and robotics is recommended.

Learning Objectives:

At the end of this course you will have gained knowledge about applications of various topics in designing safe and interactive autonomous systems: temporal logics, reactive synthesis, planning and control, learning and human modeling, game theoretical foundations of interactive systems, safe learning, etc.

You will also have hands-on experience working on a research project and it is expected that you will gain the following research skills: analyzing literature related to a particular topic, critiquing papers, and presentation of research ideas.

Staff

Dorsa Sadigh

Instructor

Office Hours: Mondays 11am-12pm, Gates 142 (by appointment)

Webpage

Toki Migimatsu

Course Assistant

Timeline

Date	Lecture	Handouts / Deadlines	Notes
Tue, Sep 26	Introduction to Safe and Interactive Robotics	Syllabus	Template for Scribes Survey Slides
Thu, Sep 28	Lecture: Motion Planning	Spatial Planning: A Configuration Space Approach. Lozano-Perez. (1983). Analysis of Probabilistic Roadmaps for Path Planning. Kavraki, et al. (1988). Randomized Kinodynamic Planning. LaValle, et al. (2001). Path Planning in Expansive Configuration Spaces. Hsu, et al. (1997).	Laura, Charles Notes
Tue, Oct 03	Lecture: Trajectory Optimization		Mingyu, Maxime, Keven Notes
Thu, Oct 05	Lecture: Formal Methods in Robotics	Church’s Problem Revisited. Kupferman and Vardi. (1999). Temporal Logic-based Reactive Mission and Motion Planning. Kress-Gazit, et al. (2009). A Fully Automated Framework for Control of Linear Systems from Temporal Logic Specifications. Kloetzer, et al. (2008). Synthesis for Human-in-the-Loop Control Systems. Li, et al. (2014). Optimization-based Trajectory Generation with Linear Temporal Logic Specifications. Wolff, et al. (2014). Model Predictive Control with Signal Temporal Logic Specifications. Raman, et al. (2014).	Eli, Hesam, Kyle, Chelsea Notes
Tue, Oct 10	Lecture: Formal Methods in Robotics	Church’s Problem Revisited. Kupferman and Vardi. (1999). Temporal Logic-based Reactive Mission and Motion Planning. Kress-Gazit, et al. (2009). A Fully Automated Framework for Control of Linear Systems from Temporal Logic Specifications. Kloetzer, et al. (2008). Synthesis for Human-in-the-Loop Control Systems. Li, et al. (2014). Optimization-based Trajectory Generation with Linear Temporal Logic Specifications. Wolff, et al. (2014). Model Predictive Control with Signal Temporal Logic Specifications. Raman, et al. (2014).	Karen, Vikranth Notes
Thu, Oct 12	Reading: Safe Learning and Control	P1: A General Safety Framework for Learning-Based Control in Uncertain Robotic Systems. Fisac, et al. (Pros: Hesam, Cons: Sumeet). Concrete Problems in AI Safety. Amodei, et al. (2016). Reachability-based Safe Learning with Gaussian Processes. Akametalu, et al. (2014). Safe Exploration for Optimization with Gaussian Processes. Sui, et al. (2015). Safe Control under Uncertainty with Probabilistic Signal Temporal Logic. Sadigh, et al. (2016). Safe Visual Navigation via Deep Learning and Novelty Detection. Richter, et al. (2017). Diagnosis and Repair from Signal Temporal Logic Specifications. Ghosh, et al. (2016).	Sumeet, Hesam
Tue, Oct 17	Reading: Safe Learning and Control	P2: Bayesian Optimization with Safety Constraints: Safe and Automatic Parameter Tuning in Robotics. Berkenkamp, et al. (Pros: Lin, Cons: Hans). P3: Safe Visual Navigation via Deep Learning and Novelty Detection. Richter, et al. (Pros: Mingyu, Cons: Maxime).	Mingyu, Lin, Maxime, Hans
Thu, Oct 19	Reading: Adversarial Neural Networks	P1: Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images. Nguyen, et al. (Pros: Nipun, Cons: Kyle). P2: Explaining and Harnesssing Adversarial Examples. Goodfellow, et al. (Pros: -, Cons: -). P3: Understanding Neural Networks Through Deep Visualization. Yosinski, et al. (Pros: Pengda, Cons: Shane). Exploring the Space of Adversarial Images. Tabacof, et al. (2015). The Limitations of Deep Learning in Adversarial Settings. Papernot, et al. (2016). Adversarial Manipulation of Deep Representations. Sabour, et al. (2015).	Nipun, Kyle, Shane, Pengda
Tue, Oct 24	Reading: Models of Cognition	Project Proposal Reports Due P1: Probabilistic Models of Cognition: Exploring representations and inductive biases. Griffiths, et al. (2010).(Pros: Chelsea, Cons: Yuhang) . P1_A: Probabilistic Models of Cognition: Conceptual Foundations. Chater, et al. (2006). P1_B: Prospect Theory. Kahneman and Tversky. (1979). P2: Helping People Make Better Decisions Using Optimal Gamification. Lieder, et al. (2016). (Pros: Pengda, Cons: Sharon). P2_A: A Reward Shaping Method for Promoting Metacognitive Learning. Lider, et al. (2016).	Sharon, Chelsea, Yuhang, Pengda
Thu, Oct 26		Project Proposal Presentations
Tue, Oct 31		Project Proposal Presentations
Thu, Nov 02		Project Proposal Presentations
Tue, Nov 07	Lecture: Learning from Demonstration	Maximum Margin Planning. Ratliff, et al. (2006). Maximum Entropy IRL. Ziebart, et al. (2010). Movement Primitives via Optimization. Dragan, et al. (2015). Active Preference-Based Learning of Reward Functions. Sadigh, et al.(2017).	Lin, Gleb
Thu, Nov 09	Reading: Learning from Demonstration	P1: Socially Compliant Mobile Robot Navigation via IRL. Kretzschmar, et al. (2016). (Pros: Sumeet, Cons: Karen) . P2: Predicting Human Reaching Motion in Collaborative Tasks using Inverse Optimal Control and Iterative re-planning. Mainprice, et al. (2015). (Pros: Sharon, Cons: Hans). Trajectories and Keyframes for Kinesthetic Teaching. Akgun, et al. (2012). Designing Robot Learners that Ask Good Questions. Cakmak, et al. (2012). Using perspective taking to learn from ambiguous demonstrations. Breazeal, et al. (2012). Understanding Intentions of Others. Meltzoff, et al. (1995). Infant Imitation After a 1-Week Delay. Meltzoff, et al. (1988). Rational Imitation in Preverbal Infants. Gergely, et al. (2002). Planning Based Prediction for Pedestrians. Ziebart, et al. (2009).	Karen, Sumeet, Hans, Sharon
Tue, Nov 14	Guest Lecture: Mo Chen	Project Milestone Reviews Due
Thu, Nov 16	Reading: Intent Inference	P1: Shared Autonomy via Hindsight Optimization. Javdani, et al. (2015). (Pros: Yuhang, Cons: Gene) P2: Generating Legible Motion. Dragan, et al. (2013). (Pros: Gleb, Cons: Haruki) Goal Inference as Inverse Planning. Baker, et al. (2007). Intention-Aware Motion Planning. Bandyopadhyay, et al. (2013). Manipulating Mental States through Physical Action. Gray, et al. (2014). Generating Plans that Predict Themselves. Fisac, et al. (2016). Expressing Thought: Improving Robot Readability with Animation Principles. Takayama, et al. (2011). Anticipation in Robot Motion. Gielniak, et al. (2011). Robot Navigation in Dense Human Crowds. Trautman, et al. (2013). Planning for Autonomous Cars that Leverages Effects on Human Actions. Sadigh, et al. (2016). Information Gathering Actions over Human Internal State. Sadigh, et al. (2016). Robot Planning with Mathematical Models of Human State and Action. Dragan. (2017). Formalizing Human-Robot Mutual Adaptation: A Bounded Memory Model. Nikolaidis, et al. (2016).	Yuhang, Gleb, Gene, Haruki
Tue, Nov 21	Thanksgiving Break
Thu, Nov 23	Thanksgiving Break
Tue, Nov 28	Reading: Communication and Coordination	P1: Inverse Reward Design. Hadfield-Menell, et al. (2017). (Pros: Vikranth, Cons: Karen) Learning Behavior Styles with Inverse Reinforcement Learning. Lee, et al. (2010). Knowledge and implicature: Modeling language understanding as social cognition. Goodman, et al. (2013). Coordination Mechanisms in Human-Robot Collaboration. Mutlu, et al. (2013). Conversational Gaze Aversion for Humanlike Robots. Andrist, et al. (2014). Robot Deictics: How Gesture and Context Shape Referential Communication. Sauppé, et al. (2014).	Karen, Vikranth
Thu, Nov 30	Reading: Collaboration	P1: Cooperative Inverse Reinforcement Learning. Hadfield-Menell, et al. (2016). (Pros: Shane, Cons: Nipun) P2: Asking for Help Using Inverse Semantics. Tellex, et al. (2014). (Pros: Kyle, Cons: Mingyu) Implicitly Assisting Humans to Choose Good Grasps in Robot to Human Handovers. Bestick, et al. (2016). Human-Robot Cross-Training. Nikolaidis, et al. (2013).	Shane, Nipun, Mingyu, Kyle
Tue, Dec 05	Project Presentations
Thu, Dec 07	Project Presentations
Tue, Dec 12	Project Reports	deadline at midnight (firm)

Grading

Component	Contribution to Grade
Final Project	50%
Student Presentations & Paper Reviews	30%
Scribing & Class Participation	20%
Total	100%

Project Grading

Component	Contribution to Grade
Project Proposal Reports	5%
Project Proposal Presentations	5%
Project Milestone Reviews	10%
Project Presentation (possibly with demos)	15%
Final Project Report	15%
Total	50%