August 20, 2019 by Tyler Charboneau
Sundance Multiprocessor Technology has introduced the VCS-1 processor, available as a comprehensive system that blends advanced hardware and software capabilities. The stack was specifically designed for applications involving VCS: vision, control, and sensors—per its namesake.
The VCS-1 processor was created in collaboration with the VineScout Project, an initiative focused on camera-based data collection. The product’s corresponding system-on-chip (SoC) also played a key role in operating the Robotnik Summit-XL.
Universally, the VCS-1 (pictured below) flaunts its ability to effectively coordinate data capture and navigation duties in autonomous systems. It was developed in response to the growing ubiquity of sensor technology.
Sundance’s VCS-1 processor, showing board layout and I/O. Image courtesy of Sundance.
The VCS-1 is classified as an SoC, comprised of two key components working in harmony. It’s a PCI/104 Linux stack, with an EMC2 board and an FM191 expansion card. This EMC2 board is a PCIe/104 OneBank carrier, configured for Trenz-compatible modules.
Meanwhile, the expansion card provides ample ADC, DAC, and I/O. These grant the processor flexibility and support for outside connections. Such options are ideal for sensors and associated hardware. PCI/104 is popular within embedded systems due to its performance and scalability.
OneBank’s advantages include free PCB real estate and stack flexibility. Engineers have options when designing board layouts, and can incorporate a greater number of components. Additionally, OneBank uses a smaller bus connector that’s relatively low cost. Despite its stature, this connector meshes well with full-sized PCIe/104 connectors that are widely used today.
OneBank incorporates only 1 of 3 PCIe/104 connector banks, resulting in a 52-pin—rather than the traditional 156-pin—configuration, which delivers the functionality of traditional connections with impressive bandwidth. As a bonus, it also offers potential cost savings.
Another view of the VCS-1, showing additional features. Courtesy of Sundance.
Trenz modules can supply additional processing resources, increased I/O availability, and power regulation. These are indispensable when working with complex applications. Trenz modules are regarded for their heat dispersion and compact design.
The VCS-1 incorporates a Xilinx Zynq UltraScale+ MPSoC at its core. This adds a quad-core Arm Cortex-A53 platform into the equation, which is 64bit. This implementation excels at real-time control with soft and hard engines, characteristic of graphics and video applications.
The VCS-1 is renowned for its compatibility with advanced camera systems, making this extremely valuable. Accordingly, the A53 supports various video codecs. Xilinx MPSoCs boast diverse hardware and software support, making them ideal for the VCS-1.
This is paired with an Arm Mali 400 GPU and an Arm Cortex-R5 real-time processing unit, (aka RPU). The R5 supports SoC integrations, internal safety controls, and error management.
The said components cooperate well on a physical level. They’re compact and pack impressive performance into small packages—key drivers behind the VCS-1’s design. Sundance’s SoC is small and lightweight as a result. Robotics vary greatly in size, which necessitates the development of a space-friendly processor. For example, the PCI/104 measures a mere 90 x 96 mm.
While many will know that the VCS-1 specialises in autonomous navigation and video capture, the processor is also a powerful deep learning and EDGE AI platform. The system leans heavily on its Xilinx partnership, bringing support for the Xilinx reVision stack and Deep Learning Processor Unit (DPU). This DPU integrates programmatically with the UltraScale+ SoC, allowing engineers to tap into neural-networking capabilities. ReVision focuses on computer vision and machine learning, allowing the VCS-1 to process complex operations simultaneously. This productivity advantage comes via low latency and easy software integrations, according to the user’s specific needs.
The VCS-1 supports many popular neural networks, including YOLO, AlexNet, GoogLeNet, CAFFE, DarkNet, TensorFlow VGG, SSD, and FCN. Custom neural network (aka CNN) builds require the Xilinx DNNDK (Deep Neural Network Development Kit).
Of course, hardware demands in robotics vary widely depending on complexity; thankfully, the processor’s hardware support is seemingly endless thanks to its said, massive variety of connectors.
The European Union continually sponsors technology projects, which aim to make meaningful impacts in diverse industries. Across different markets, these programmes favour breakthrough products, services, and more. The Sundance VCS-1 is an approved development of this FTI initiative.
The globe is embracing automation across multiple markets now more than ever, and the VCS-1 is a key cog in that transition. Due to its versatility and performance, we expect Sundance’s latest innovation to be a hit technology with engineers in emerging fields.
