Masters Thesis - Learning from Demonstration applied to Poppy

Hi everyone,

I am a masters mechatronic student who has to do a project for his masters thesis.
My idea is to develop a Learning from Demonstration (LfD) to use on Poppy.

I think it could be useful to get closer users without programming skills to robots in order to use them for every kind of task (eg houseworks for everyday uses; “personalized” rehabilitation - something with Poppy was already done in that sense using it as “teacher” [1], I thing it is possible to use it also as physiotherapist [2] -; high-mix production - industrial field - etc… )

I think the open-source nature of Poppy, with the LfD, could be the key to increase the dissemination and use of robotics by anyone.

Premising that my skills are those of a student and that I want to learn everything necessary, do you know someone could be interested to develop something like that? (I am ready to reach anyone from anywhere in the world for as long as necessary!)



[2] M. Maaref, A. Rezazadeh, K. Shamaei, M. Tavakoli “A Gaussian Mixture Framework for Cooperative Rehabilitation Therapy in Assistive Impedance-based Tasks” Aug. 2016.

LBD is definitely something interesting to use robots without any programming skills.
I’m not sure to understand your question or if you have one, but @damiend and @Mai are working (worked ?) for the Keraal project you are mentioning.

Also, I played with Gaussian Mixture Models for lbd some years ago (during my master internship), to mix multiple recordings of a movement and generalize it. You can find what I did here.
I really suggest you this book from Sylvain Calinon which is a very good reference for lbd.


thanks for your reply.

I try to be more clear: since I have to write a thesis on a project and I would like to (learn to) work on LbD applied to Poppy, is it possible to work on such a project? Specifically, do you know someone who is already working on a similar project? Or do you know anyone who would like the idea of starting such a project?

I will definitely take a look to this material, thanks again!


sorry I missed out this post, and only saw it today.

For recording movements by kinesthetics, Poppy is indeed very convenient. The usual library offers a good API for this, You can also find a user interface to record and combine movements and create an exercising session on (the first part of the video shows how it works)

For using Poppy in the physiotherapy context, you can check our projet Keraal on
We used Gaussian Mixture Models for movement analysis. You can refer to the article :

Devanne, M. and Nguyen, S. M. (2017). Multi-level Motion Analysis for Physical Exercises Assessment in Kinaesthetic Rehabilitation, IEEE International Conference on Humanoid Robots (Humanoids),

For personalisation, we have worked on how to adapt the motions to patient’s limits (for each articulation) , using Gaussian Mixture Models and Gaussian Process in :

Maxime Devanne, Sao Mai Nguyen. Generating Shared Latent Variables for Robots to Imitate Human Movements and Understand their Physical Limitations. Leal-Taixé L; Roth S. Lecture Notes in Computer Science , 11130, Springer, Cham, pp.190-197, 2019, Computer Vision – ECCV 2018 Workshops, ⟨10.1007/978-3-030-11012-3_15⟩. ⟨hal-01891414v3⟩