Biologically-Inspired Learning for Humanoid Robots

Lecturer (assistant)
Number0000002585
Type
Duration4 SWS
TermSommersemester 2015
Language of instructionEnglish
Position within curriculaSee TUMonline
DatesSee TUMonline

Dates

Course criteria & registration

Objectives

"The lectures will provide a detailed presentation of humanoid robots learning with practical applications on a humanoid robot platform. The class make connection with the experimentation on a humanoid robot, students will be able to investigate learning models on real robots. At the end of the module, students will be able to create their own model to solve humanoid robot problems that benefits from learning. Furthermore, students will be exposed to brain regions that are involved in the learning process. "

Description

"1. Introduction 1.1. Motivation 1.1.1. Humanoids robots 1.1.2. Legged robots 1.1.3. Human tasks and humanoids 1.1.4. Humanoids problems (walking, navigation, manipulation, stirs, balancing, grasping…) 1.2. Human brain 1.2.1. Neurons 1.2.2. Loops 1.2.3. Learning strategies 1.3. Learning and development 1.4. Tutorial1: familiarization with the simulator (Webots) 2. Humanoid learning overview 2.1. What humanoid need to learn? 2.2. Supervised and unsupervised learning 2.3. Learning by self-exploration 2.4. Learning by demonstration 2.5. Tutorial2: first step with a humanoid robot (NAO) 2.5.1. Example in simulation (NAO humanoid robot in Webots) 2.5.2. Example on NAO humanoid robot. 3. Humanoid supervised learning 3.1. Learning rules 3.1.1. Delta rule 3.1.2. Back propagation 3.2. Neural networks 3.3. Kernel-based techniques 3.4. Error-based learning in the cerebellum (cerebellar model articulation controller- CMAC) 3.5. Tutorial3: comparator model of the cerebellum (balance control during standing position) 4. Humanoid unsupervised learning 4.1. Learning rules 4.1.1. Hebbian learning 4.1.2. Competitive learning 4.2. Neural networks 4.2.1. Self-organising maps 4.2.2. Hopfield networks 4.3. K-means clustering 4.4. Unsupervised learning in the cortex 4.5. Tutorial5: learning of a motor map for humanoid robot reaching 5. Humanoid reinforcement learning 5.1. Introduction 5.2. Methods 5.2.1. Temporal difference methods 5.2.2. Q-learning 5.2.3. Qualitative adaptive reward learning QARL 5.3. Reinforcement learning in the basal ganglia 5.4. Tutorial6: action selection for humanoid robot walking in different conditions (e.g. terrain slops) 6. Humanoid transfer of learning 6.1. A background: human inspiration 6.2. What to transfer? 6.3. Transfer with different conditions 6.4. Transfer with different tasks 6.5. Tutorial8: humanoid robot draws basic forms on vertical boards based on his experience in drawing on horizontal one. 7. Humanoid grasping and manipulation 7.1. Vision-based grasping 7.2. Object recognition and localization 7.3. Bimanual grasping 7.4. Tutorial9: humanoid robot opening and closing door 8. Humanoid robots: perspectives "

Prerequisites

C/C++ programming skills, MATLAB, mathematics. Prior knowledge about artificial neural networks is highly recommended

Teaching and learning methods

"The lectures will provide a detailed presentation of humanoid robots learning with practical applications on a humanoid robot platform. The class make connection with the experimentation on a humanoid robot, students will be able to investigate learning models on real robots. At the end of the module, students will be able to create their own model to solve humanoid robot problems that benefits from learning. Furthermore, students will be exposed to brain regions that are involved in the learning process. "

Examination

"laboratory assignments and presentation, oral exam. Each tutorial session is composed of one exercise, which students have to solve in groups (2 students per group). We offer 2 blocks of exercises, which students have to solve it individually. At the end of the module each student (or group) presents a relevant research paper from the related literature in the field. The final grade is composed as follows: A) final oral exam: 30% B) laboratory assignments : 30% C) individual laboratory assignments : 30% D) paper presentation: 10% "

Recommended literature

-

Links