[Paper Review] Robust robotic control on the neuromorphic research chip Loihi.

Carlo Michaelis, Andrew B. Lehr+1 more|arXiv (Cornell University)|Aug 26, 2020

Advanced Memory and Neural Computing4 citations

TL;DR

This paper presents a neuromorphic control algorithm based on the anisotropic spiking neural network model, implemented on Intel's Loihi chip to enable robust, low-latency robotic arm control. It successfully maps millisecond-scale spiking activity to control-relevant timescales, generating stable multidimensional trajectories with high variability and noise resilience.

ABSTRACT

Neuromorphic hardware has several promising advantages compared to von Neumann architectures and is highly interesting for robot control. However, despite the high speed and energy efficiency of neuromorphic computing, algorithms utilizing this hardware in control scenarios are still missing. One problem is the transition from fast spiking activity on the hardware, which acts on a timescale of a few milliseconds, to a control-relevant timescale on the order of hundreds of milliseconds. Another problem is to enable the execution of complex trajectories, requiring the spiking activity to contain sufficient variability, while at the same time, for reliable performance, network dynamics require adequate robustness against noise. In this study we exploit a recently developed biologically-inspired spiking neural network model, the so-called anisotropic network, as the basis for a neuromorphic algorithm for robotic control. For this, we identified and transferred the core principles of the anisotropic network to neuromorphic hardware using Intel's neuromorphic research chip Loihi and validated the system on trajectories from a motor-control task performed by a robot arm. We show that the anisotropic network on Loihi reliably encodes sequential patterns of neural activity, each representing a robotic action, and that the patterns allow the generation of multidimensional trajectories on control-relevant timescales. Taken together, our study presents a new algorithm that allows the control of complex robotic movements using state of the art neuromorphic hardware.

Motivation & Objective

• To address the challenge of translating fast neuromorphic spiking activity (ms timescale) into stable, control-relevant motor commands on the 100-ms timescale.
• To enable complex, multidimensional robotic trajectories using spiking neural networks on neuromorphic hardware.
• To ensure robustness against noise while maintaining sufficient variability in neural patterns for diverse motor actions.
• To validate the approach on real robotic control tasks using a physical robot arm.

Proposed method

• Adapted the biologically inspired anisotropic network model for implementation on Intel's Loihi neuromorphic chip.
• Mapped sequential neural activity patterns to discrete robotic actions via spike timing and rate coding.
• Engineered network dynamics to sustain stable activity patterns over hundreds of milliseconds despite inherent noise.
• Used spiking neuron dynamics with synaptic plasticity to encode and replay temporal sequences of motor commands.
• Calibrated network parameters to balance variability for trajectory diversity and robustness for reliable performance.
• Validated the system using real-world robotic motor control tasks with a physical robot arm.

Experimental results

Research questions

• RQ1 Can neuromorphic hardware like Loihi reliably encode and generate control signals for complex robotic trajectories?
• RQ2 How can fast spiking activity (ms timescale) be translated into control actions on the 100-ms timescale required for robotic motion?
• RQ3 To what extent can anisotropic spiking networks maintain robustness against noise while supporting diverse motor patterns?
• RQ4 Can the network generate stable, multidimensional trajectories without external feedback or explicit timing signals?

Key findings

• The anisotropic network on Loihi successfully encoded sequential neural activity patterns representing distinct robotic actions.
• The system generated stable, multidimensional trajectories on control-relevant timescales (hundreds of milliseconds) from fast spiking dynamics.
• Robustness against noise was achieved without compromising the variability needed for complex trajectory generation.
• The algorithm enabled reliable execution of complex motor tasks on a physical robot arm using only spiking neural network dynamics.
• The approach demonstrated energy efficiency and low-latency control, leveraging the inherent advantages of neuromorphic hardware.
• The transfer of biologically inspired network principles to neuromorphic hardware enabled scalable and adaptive robotic control.