Seminar "Reinforcement Learning in Autonomous Driving"

Organizer: M.Sc. Sina Shafaei

Contact: shafaei (at) in.tum.de

Modul: IN2107

Registration: Via Matching System

Type: MasterSeminar

Semester: SS 2018

ECTS: 5.0/4.0

Time & Location: 02.09.023 / 13:00-15:00

                                                                            


News


  • The first session of the seminar (details, topic assignments, team members) will be on 20.04.2018 at 02.09.023, 13:00-15:00

  • Registration is only via Matching System, 09.02-20.02, and there is no need to submit any document (C.V, motivation letter, etc.) beforehand

Content

Reinforcement Learning (RL) is a kind of learning that allows autonomous agents to learn using feedback received from the environment. The basic idea is inspired by nature itself, based on the manner that people and animals learn. RL is based on trying actions and observing what happens in the environment. If actions lead to better situations, there is the tendency of applying such behavior again, otherwise, the tendency is to avoid such behavior in the future. Therefore, the problem is reduced to learn how to select optimal actions to be performed in each situation to reach a given goal.

This approach of machine learning, implies to acquire new knowledge to improve the performance of an agent interacting with its environment. However, the agent is not told what actions to take. The agent has to discover by itself what actions lead to more reward by trial and error.

In this seminar, students will investigate the role and the future of the Reinforcement Learning on different scenarios and use cases which are important in autonomous driving and can have a great impact on identifying the research direction in this field.

 

 


Workflow

In our first session (20.04.2018) we will distribute the topics among the groups of students (groups of two members). Groups will collect the necessary materials (state of the art scientific works like publications) within the context of their topic and will perform their research based on the collected resources. Each group will be given 40 minutes on their specified date (are listed in below) to represent the results and afterwards will write and submit a scientific paper (IEEE template) with minimum 6 pages, explaining the contribution of their work in this seminar.


                                  TopicsPresentation (Date)
A- RL-based Use Cases in Autonomous Driving, Current Status, Vision, Challenges29.06.2018
B- Transfer/Deep/Reinforcement Learning and Challenges29.06.2018
C- Apprenticeship and Interactive Reinforcement Learning in Autonomous Driving06.07.2018
D- Safety of the RL-based Functions, Challenges, Approaches06.07.2018
E- Safe Reinforcement Learning and the Future of ISO 2626213.07.2018
F- Learning from Human Generated Rewards13.07.2018
G- Using Behavior in Learning from Human Reward20.07.2018
H- Reward Shaping for RL by Emotion Expressions, Challenges, Approaches20.07.2018
I- Formal Verification of RL-based Approaches27.07.2018
J- Safety Limiter Architectures and Frameworks for RL-based Functions27.07.2018
K- Federated Control of Reinforcement Learning Agents in Autonomous Driving 03.08.2018
L- Reinforcement Learning and Ethical Issues03.08.2018

Material