Embedded processor module paves way for precision robotics
Computer vision, edge artificial intelligence (AI) and deep learning are brought together in the VCS-1 low power, embedded processor module by Sundance Multiprocessor Technology.
The VCS-1 is designed specifically for precision robotics incorporating complex, real-time vision, control and sensor applications.
It has been developed and proven as part of the European Union’s H2020 ‘Fast-Track-Innovation’ pilot programme. The VineScout (VS) uses robotics to collect real-time data from vineyards to devise grape maturation and harvesting strategies.
The PC/104 form factor measures 90 x 96mm and provides industry-standard compatibility and expandability, says Sundance. The VCS-1 embedded processor module is optimised for computer vision, edge artificial intelligence (AI) and deep learning requirements. It weighs just 300g, has a low power consumption of typically 15W and is compatible with a range of commercially available sensors and actuators.
The Sundance VCS-1 is based on a Xilinx Zynq MPSoC, mounted on to the PC/104 board using a system on module (SoM). There is also an Arm Cortex A53 64-bit quad-core processor which combines real time control through engines for graphics, video, waveform, and acceleration with an FPGA. Characteristics include an Arm Mali 400 graphics processing unit (GPU) for graphics acceleration, an Arm Cortex R5 real time processing unit (RPU) for managing real time events and the programmable FPGA logic for hardware acceleration of AI algorithms used for on-the-fly image processing.
The Sundance VCS-1 features extensive I/O capabilities available through the Sundance external interface card (SEIC), including multiple USB3 interfaces for interfacing various cameras and sensors such as the Intel RealSense T265 tracking camera, Intel RealSense D435 and Stereo Labs’ Zed depth cameras and FLIR AX-8 thermal camera. It can also connect with most Arduino and Raspberry Pi actuators and sensors. One interface enables it to mimic a PC with HDMI display, SATA storage and Ethernet networking. An on-board ADC gathers data from an external sensor and there is an on-board DAC to control devices such as servos. I/O standards are also implemented directly on the programmable logic to reduce the latency between the various supported cameras, sensors and servos.
Software support for precision robotics includes the ROS Melodic Morenia (ROS- compatible and ROS2-ready) robotics platform, MQTT machine-to-machine (M2M) connectivity protocol, OpenCV computer vision library of real time programming functions, Xilinx’s Edge-AI solutions and the Python scripting language. The Ubuntu operating system, Xilinx SDSoC environment, TULIPP’s STHEM toolchain and Xilinx DPU (deep learning processing unit) for convolutional neural networks are also supported.
The Sundance VCS-1 is available in a custom enclosure. The PC/104-Blade, is designed to remove the need for any fans and provide a rugged environment for building embedded applications for resilient, precision robotics.
The Sundance VCS-1 precision robotics platform has been developed in conjunction with the VineScout viticulture partners, which includes the French Agri-robotics manufacturer, Wall-YE.
The project’s aim is to develop a robot for vineyard monitoring to help wine producers measure key parameters of their vineyards, including water availability, the temperature of the leaves and plant robustness. Capable of operating 24/7, the VineScout has a battery life of six hours and provides winemakers with real time data on vine and grape growth and maturation, so that they can optimise irrigation and harvesting.