This work was funded by the Sao Paulo Research Foundation (FAPESP), grants 2017/02377-5, 2017/01687-0 and 2018/25902-0, and the Neuro4PD project - Royal Society and Newton Fund (NAF\R2\180773). Moioli acknowledge the support from the Brazilian institutions: INCT INCEMAQ of the CNPq/MCTI, FAPERN, CAPES, FINEP, and MEC. This research was carried out using the computational resources from the CeMEAI funded by FAPESP, grant 2013/07375-0. Additional resources were provided by the Robotics Lab within the ECR, and by the Nvidia Grants program.