This is a guest post from Flemming Christensen, Managing Director of Sundance Multiprocessor Technology Ltd.
As a Xilinx Alliance Partner, Sundance, recently launched the VCS-1 Platform.
The VCS-1 is a COTS solution, based on industry-standard PC/104 form-factor, and GitHub open-sourced Zynq MP-SoC solution that can to control and navigate any ROS-enabled robot, like in this video.
The VCS-1 Platform is targeted at robotics application that require low-power and low-latency. It provides the optional ability to accelerate specific AI algorithms, like classifying water-stress levels in plants and detecting any abnormalities in growth. The VCS-1 was used in a robot for vineyard monitoring, as part of the VineScout project. The VCS-1 replaced a conventional CPU solution that took more than 150W with an equally powerful processing solution using only 15W.
The choice of the Xilinx Zynq SoC, rather than using a GPU to accelerate the AI alongside the CPU, was evaluated during an R&D project called Tulipp. One of many benefits of choosing Xilinx were the SDSoC tools, allowing design in “C” that results in custom accelerators being implemented in the Zynq’s FPGA fabric. Another benefit was the Xilinx PYNQ/Python environment for easy control of I/O. Moving forward we are excited about the possibilities enabled by the new Vitis tools that Xilinx recently announced.
The arrival of Xilinx’s DPU to provide Convolutional Neural Network has been a major boost to the VCS-1 platform, enabling AI for object detection (trees, fruit, etc.) to be performed by the robot at the edge.
The VCS-1 has the following technical specifications:
Quad ARM-A53 Zynq integrating H265 codec, using the ZU4EV MP-SoC
15x Digital I/O Ports;12x Analog Inputs; 8x Analog Outputs on DB9
4x USB3.0, SATA, HDMI & USB2 to work as a Linux Single-Board Computer
Raspberry Pi I/O Header for 28x I/O GPIOs
ROS, OpenCV, Xilinx DPU/SDSoC, MQTT and PYNQ/Python software support