INRC2

AI embodiment over 3D simulation and neuromorphic hardware

INRC2

The embodiment of artificial intelligence (AI) is a key to further adoptance of neuromorphic computing in real world applications. The Neurorobotics Platform 3D simulation tool and Intel‘s Loihi research chip combine their strengths to enable it, by providing body and brain to robotic artefacts.

Project description

The Neurorobotics Platform (NRP) is a web-based software co-developed by fortiss in the frame of the Human Brain Project, that enables neuroscientists and AI experts to connect spiking neural networks with simulated bodies, to bridge the gap between AI algorithms and real-world applications, to give AI algorithms applications in real-world use cases.

We first have integrated Intel‘s research neuromorphic chip Loihi into the NRP over a well established spiking neural network simulator called Nengo and to use it as a benchmarking tool to compare the execution and learning of spiking neural networks over different simulators, including Nest, SpiNNaker, Nengo and of course Loihi. The benchmarking experiment features a mobile swimming robot controlled with a spiking Central Pattern Generator (CPG). This has based on previous work in project INRC1.

In a second phase, we have integrated Loihi directly into the NRP, without Nengo, in order to run faster control algorithms in simulation with Loihi-in-the-loop. This enabled us to test a simulated robotic arm trained in the NRP with Reinforcement Learning for object insertion, in the frame of INRC3.

Research contribution

This project was mostly about engineering as the actual integration of the Loihi research chip into the NRP. Though, the scientific impact is huge, as it enables other research projects. Having neuromorphic hardware in the simulation loop is essential to experiment and research development.

fortiss has a key expertise in implementing the middleware to enable running spiking neural networks in Loihi over Nengo or directly in realistic 3D simulated experiments.

The research enabled by this project, namely research in robot locomotion and control, is very important because it is an emerging, yet intensively investigated field. We have successfully shown that this is possible in simple embodiments, like snakes and their oscillatory locomotion, or robotic arm with reinforcement learning.

Project duration

01.11.2019 - 31.03.2022

Intel and the Intel logo are trademarks of Intel Corporation in the U.S. and other countries.

Further projects

INRC1

Nonlinear Spiking Locomotion In Fluids With Neuromorphic Hardware

Spiking robot control with neuromorphic hardware is a break through in energy efficient AI driven robot control. It allows for mobile robots, like the…
INRC3

Neuromorphic Manipulator-Arm Controller For Object Insertion With Force Feedback

In this project, we are implementing object insertion with a robotic arm and spiking neural network based reinforcement learning with force (haptic) feedback.
Human Brain Project

Bridge the gap between neuroscience computing and robotics

Equipping robots with artificial neural networks (NN) offers major advantages. They are not only easier to control. When furnished with sensors, they can also…

fortiss Lab

Contact

More information

Project partner

click to return to top of page