Humans and animals seamlessly control their complex bodies, use new tools, and adapt to injuries.
A highly adaptive body model of some sort (body schema/body image, forward and inverse models, etc.) seems indispensable.

Using the Mujoco simulator we simulate infant-like full-body tactile agents such as iCub and MIMo environment. We have developed an SNN that learns, in an unsupervised manner, the association between the proprioceptive state (joint angles) of the agent and the corresponding 
double-touch activations that occurred during random motor exploration. 
We use coordinate transformation [Pouget-1997], gain modulation mechanisms [Silver-2011], Spike timing-dependant plasticity (STDP). The model learned how to reach its own body during motor babbling.
A non-neuromorphic model already exists [Marcel-2022] and will serve as a baseline (see video https://youtu.be/gg_qySTwOdc?si=XptE8oI-XysICbp0).  

In this workgroup, we propose to develop the closed-loop end-to end interface between the SNN model and the robot/robotic simulator

To this end, we will try to:
1- Deploy the SNN on SpiNNaker2 and other platforms.
2- Connect the neuromorphic platform to the physical simulator and 2 DoF Manipultor and stream data in and out.
3- Investigate other possible SNNs and learning algorithms to learn embodiment.

Fulfilling these tasks will allow the community to advance on these topics:

Closed-loop end-to-end platform to learn embodiment and body schema development
A platform for developing reinforcement learning algorithms on humanoid robotic simulators with SNNs .
A bio-inspired model of body schema development.

To get an idea of this workgroup visit our repository where you can find explanations and an interactive model.

https://github.com/event-driven-robotics/LEARNING-NEUROMORPHIC-BODY-SCHEMA