[PhD topic] Meta-learning for adaptive whole-body control
Contact / PhD advisor: firstname.lastname@example.org
The LORIA recently received a TALOS robot, which is a full-scale, torque-controlled humanoid (http://pal-robotics.com/robots/talos/). We are currently assembling a team at the intersection of model-based control and machine learning to make it move as well as possible, and, in particular, to adapt to new situations.
Current humanoid robots use whole-body controllers that leverage a model of the dynamics (or the kinematics) of the robot and online optimization . The objective of this PhD will be to learn to correct this model using data acquired by the robot. To learn this model quickly, we will leverage recent advances in meta-learning . The main challenges are: (1) how to quickly learn an accurate model for a humanoid robot? (2) and how to integrate learned models in the model-based control loop.
 Escande A, Mansard N, Wieber PB. Hierarchical quadratic programming: Fast online humanoid-robot motion generation. The International Journal of Robotics Research. 2014 Jun;33(7):1006-28.
 Finn C, Abbeel P, Levine S. Model-agnostic meta-learning for fast adaptation of deep networks. InProceedings of the 34th International Conference on Machine Learning-Volume 70 2017 Aug 6 (pp. 1126-1135). JMLR. org.
 Nagabandi A, Clavera I, Liu S, Fearing RS, Abbeel P, Levine S, Finn C. Learning to adapt in dynamic, real-world environments through meta-reinforcement learning. arXiv preprint arXiv:1803.11347. 2018 Mar 30.
Duration: 3 years
Starting date: between Oct. 1st 2020 and Jan. 1st 2023
How to apply:
Send the following documents to email@example.com
– a motivation letter.
No offers are available for now.