Skip to content

Learning for Control

PDF icon Download in PDF

  • Lecturer: SIGAUD Olivier ()
  • Course code:
  • Student workload: 10 h of lectures, 20 h of labs
  • Credits: 3 ECTS
  • Specialization tracks: ROBOT (elective) SI MeDH
  • Semester offered: S1 S2 S3 S4
  • Language of instruction: French English
  • Targeted audience: Eng. Sc. department Other : ENSAM
  • Localization : PMC Campus Other :

Course Overview

Fundamental machine-learning tools for use in control applications, particularly robotics. Special emphasis is placed on supervised learning of robotic models from data and on reinforcement learning as a data-driven approach to optimal control.

Mots-clés : Machine learning, supervised learning, reinforcement learning

Prerequisites

Students should have previously acquired the following prerequisites to follow this course:

  • programmation python
  • Gradient calculation, statistics (expectation, variance, etc.)

Intended Learning Outcomes

By the end of this course, students will be able to:

  1. Understanding regression algorithms
  2. programmation pytorch
  3. Understanding reinforcement-learning algorithms
  4. Use of machine learning on practical simulated robotics cases

Indicative Teaching Sequence and Methods

Week C/TD/TP* Content Preparation Learn.\ outc.
S1 2 hours of lectures, 4 hours of practical work Lecture: regression, from linear least squares to regression with neural networks in PyTorch. Practical: application of regression algorithms to synthetic data
S2 2 hours of lectures, 4 hours of practical work Lecture: regression for learning robotic models—geometric, kinematic and dynamic models; forward and inverse models; examples from the literature; hybrid models combining known and learned components. Practical: learning a model from simulated data in a model-based control loop.
S3 2 hours of lectures, 4 hours of practical work Lecture: model-free reinforcement learning for policy search (feedback controllers), policy-gradient theorem, REINFORCE algorithm and the bias-variance trade-off. Practical: application of REINFORCE to a simulated system.
S4 2 hours of lectures, 4 hours of practical work Lecture: proximal policy optimization (PPO) and related algorithms. Practical: application of PPO to a simulated system.
S5 2 hours of lectures, 4 hours of practical work Lecture: model-based reinforcement learning, data efficiency, and the use of MPC and variants including TD-MPC. Practical: application of TD-MPC to a simulated system.
  • C/TD/TP respectively corresponds to lectures, tutorials and lab sessions.

Indicative Assessment of Intended Learning Outcomes (1st session)

Week Individ./group In-person/remote Type of exam Evaluated outcomes Scale %
1 TP 20%
2 TP 20%
3 TP 20%
4 TP 20%
5 TP 20%

2nde session

Session Individ./group In-person/remote Type of exam Evaluated outcomes Scale %

Bibliographic references

  • Sutton and Barto, Reinforcement learning, an introduction, MIT Press, 1998 / 2018
  • Bishop, Pattern recognition and machine learning, Springer, 2006

Logo SDI Date of generation of this unit description: 06/10/2026 Logo SDI