Robust Trajectory Generation for Robotic Control on the Neuromorphic Research Chip Loihi.

Area of Science:

• Robotics
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic hardware offers speed and energy efficiency advantages over traditional von Neumann architectures for robot control.
• Current algorithms for neuromorphic hardware in control are limited, facing challenges in timescale transitions and complex trajectory generation.
• Spiking neural networks require robust dynamics and variability for reliable performance in noisy environments.

Purpose of the Study:

• To develop and validate a new algorithm for robot control using neuromorphic hardware.
• To address the challenges of timescale transition and complex trajectory generation in neuromorphic control.
• To demonstrate the effectiveness of a biologically-inspired spiking neural network model on neuromorphic hardware.

Main Methods:

• Exploited the anisotropic network, a biologically-inspired spiking neural network model.
• Transferred the anisotropic network principles to Intel's Loihi neuromorphic research chip.
• Developed a network architecture with a pooling layer for fast spike read-out and inherent regularization.
• Validated the system on robotic arm motor-control task trajectories.

Main Results:

• The anisotropic network on Loihi reliably encoded sequential patterns of neural activity representing robotic actions.
• Generated multidimensional trajectories on control-relevant timescales using these encoded patterns.
• Demonstrated robust network dynamics suitable for reliable performance in control tasks.

Conclusions:

• Presents a novel algorithm for generating complex robotic movements on neuromorphic hardware.
• Provides a building block for advanced robotic control systems utilizing state-of-the-art neuromorphic technology.
• Highlights the potential of biologically-inspired models for efficient and effective neuromorphic robot control.