Use of MOGP for Learning Control Primitives for Robotic Arms

[AdrianPrados]

Good afternoon, my name is Adrian Prados, and I am a PhD student in Robotics at Universidad Carlos III de Madrid. My research focuses on learning from demonstration applied to robotic manipulation tasks. I am basing these techniques on Gaussian Process-based approaches, and your work related to MOGPs could be very interesting to apply to the algorithms we are developing.

We had a question regarding how to use these MOGPs. We are looking to establish relationships between several joints over time. The idea is to use multiple inputs and obtain a correlation matrix between the different joints. We are unsure if your work has the capability to provide these values directly or if a specific implementation would be required to achieve this.

The goal is to use these control primitives for a dynamic motion algorithm based on iLQR, that is, to adapt the arm movements of our robotic platform to different skills.

Thank you very much for your work, and I look forward to your response regarding these capabilities.

Un cordial saludo

[tdewolff]

Hola Adrian, Que interesante la investigación que están haciendo. Sí, se puede usar MOGPTK para esta tarea. Podrías ver este ejemplo que se parece: https://games-uchile.github.io/mogptk/examples.html?q=example_human_activity_recognition. Además, acá hay un ejemplo que genera una matriz de correlación: https://games-uchile.github.io/mogptk/examples.html?q=example_gold_oil_NASDAQ_USD Avísanos si esto te sirve o si tienes otras preguntas. Saludo cordial, Taco de Wolff Op vr 24 jan 2025 om 18:47 schreef Adrian Prados ***@***.***>:

…

Good afternoon, my name is Adrian Prados, and I am a PhD student in Robotics at Universidad Carlos III de Madrid. My research focuses on learning from demonstration applied to robotic manipulation tasks. I am basing these techniques on Gaussian Process-based approaches, and your work related to MOGPs could be very interesting to apply to the algorithms we are developing. We had a question regarding how to use these MOGPs. We are looking to establish relationships between several joints over time. The idea is to use multiple inputs and obtain a correlation matrix between the different joints. We are unsure if your work has the capability to provide these values directly or if a specific implementation would be required to achieve this. The goal is to use these control primitives for a dynamic motion algorithm based on iLQR, that is, to adapt the arm movements of our robotic platform to different skills. Thank you very much for your work, and I look forward to your response regarding these capabilities. Un cordial saludo — Reply to this email directly, view it on GitHub <#75>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABKOGHXZFQPT7GC7DIDH7X32MJ4BNAVCNFSM6AAAAABV2FUUWKVHI2DSMVQWIX3LMV43ASLTON2WKOZSHAYTAMBSGM4TSNI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[AdrianPrados]

Muchas gracias por vuestra respuesta tan rápida :) y gracias por los ejemplos. Les echaré un ojo a fondo y veré si hay forma de aplicarlas a nuestra idea. Si veo que tengo alguna consulta más en detalle con esto, me pondré en contacto con vosotros.
Muchas gracias y un cordial saludo

[tdewolff]

Muchas gracias Adrían, Ojalá lo que manda Taco te sirva. Saludos! Felipe.

…

------------------------------------------------ Felipe Tobar, PhD Senior Lecturer in Machine Learning Department of Mathematics & I-X Imperial College London https://share2.ma.ic.ac.uk/~felipe/ From: Taco de Wolff ***@***.***> Date: Saturday, 25 January 2025 at 9:38 AM To: GAMES-UChile/mogptk ***@***.***> Cc: Tobar, Felipe ***@***.***> Subject: Re: [GAMES-UChile/mogptk] Use of MOGP for Learning Control Primitives for Robotic Arms (Issue #75) This email from ***@***.*** originates from outside Imperial. Do not click on links and attachments unless you recognise the sender. If you trust the sender, add them to your safe senders list<https://spam.ic.ac.uk/SpamConsole/Senders.aspx> to disable email stamping for this address. Hola Adrian, Que interesante la investigación que están haciendo. Sí, se puede usar MOGPTK para esta tarea. Podrías ver este ejemplo que se parece: https://games-uchile.github.io/mogptk/examples.html?q=example_human_activity_recognition. Además, acá hay un ejemplo que genera una matriz de correlación: https://games-uchile.github.io/mogptk/examples.html?q=example_gold_oil_NASDAQ_USD Avísanos si esto te sirve o si tienes otras preguntas. Saludo cordial, Taco de Wolff Op vr 24 jan 2025 om 18:47 schreef Adrian Prados ***@***.******@***.***>>: Good afternoon, my name is Adrian Prados, and I am a PhD student in Robotics at Universidad Carlos III de Madrid. My research focuses on learning from demonstration applied to robotic manipulation tasks. I am basing these techniques on Gaussian Process-based approaches, and your work related to MOGPs could be very interesting to apply to the algorithms we are developing. We had a question regarding how to use these MOGPs. We are looking to establish relationships between several joints over time. The idea is to use multiple inputs and obtain a correlation matrix between the different joints. We are unsure if your work has the capability to provide these values directly or if a specific implementation would be required to achieve this. The goal is to use these control primitives for a dynamic motion algorithm based on iLQR, that is, to adapt the arm movements of our robotic platform to different skills. Thank you very much for your work, and I look forward to your response regarding these capabilities. Un cordial saludo — Reply to this email directly, view it on GitHub<#75>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABKOGHXZFQPT7GC7DIDH7X32MJ4BNAVCNFSM6AAAAABV2FUUWKVHI2DSMVQWIX3LMV43ASLTON2WKOZSHAYTAMBSGM4TSNI>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>