Background
Neuromorphic computers are an exciting new alternative to traditional computing architectures and machine learning approaches. Recent academic research shows that neuromorphic computers may be efficient processors for implementing model predictive controllers, enabling low-power hardware to control more complex systems over longer prediction horizons than traditional computers.
Approach
Our approach uses an Intel Loihi 2 processor to solve a quadratic program as part of a model predictive controller that controls the trajectory of a robot arm. This will be tested both in simulation and on representative hardware.
Accomplishments
This program was recently initiated. To date, the model predictive control is implemented in simulation, and work is underway to control the simulated arm with an emulated neuromorphic processor before moving on to real-world hardware.
