Event-Driven Encoding Algorithms for Synchronous Front-End Sensors in Robotic Platforms

Abstract

Asynchronous, event-driven sampling techniques adapt the sampling rate of sensory signals to their dynamics by effectively compressing the data with respect to synchronous, clock-driven sampling. In robotics, such techniques offer data and bandwidth reduction, together with high temporal resolution and low latency. Despite the fact that vision and auditory event-driven sensors are currently available, robots are still equipped with a plethora of other sensors that might benefit from the event-driven encoding. In this paper, we study five estimation algorithms that implement the event-driven encoding for off-the-shelf clock-driven sensors. Digital accelerometer datasets were used to validate the system in robotic applications; other datasets have been used to assess the general performance of the proposed approach. The two best algorithms in terms of six performance parameters have been implemented on a Xilinx Artix-7 FPGA platform, using 2892 LUTs and 3620 flip-flops and reducing the output bandwidth from -44% to -75%, over the considered datasets.