John Deere Electronics is making its Vision Processing Unit (VPU) available to European machine builders and OEMs. Companies can now integrate this field-tested vision-based autonomy hardware into next-generation mobile machinery for precise, real-time decision-making in the field.
Computer vision and machine learning (CVML) needs high-compute solutions to analyze surroundings and respond quickly. The John Deere VPU enables equipment to respond instantly, without relying on cloud-based processing.
Developed to meet the demands of real-world outdoor applications, the VPU evolved from John Deere’s work in the agricultural space and is designed specifically for use in harsh and demanding environments. Previously exclusive to John Deere’s See & Spray technology, it is now available to OEMs for use in a range of on and off-highway mobile machinery applications. From construction, mining, robotics and material handling to on-highway and specialty vehicles, forestry and industrial applications, the flexibility of the VPU enables OEMs to tailor solutions to their specific use cases.
The VPU platform is ideal for autonomous navigation, situational awareness, and obstacle avoidance, delivering safety, efficiency and productivity in the field. Its high GPU- and high CPU-capable architecture handles multiple high-resolution camera inputs simultaneously, consolidating large volumes of visual data on machine in real-time. Interfacing through standard protocols, including CAN and Ethernet, it enables advanced algorithms and machine learning models to assess conditions and trigger automated actions as equipment is in operation. John Deere Electronics has also developed its own gigabit Ethernet switch to provide fast and seamless switching between cameras.
Crucially, the VPU has been engineered specifically for harsh, outdoor environments, where many conventional vision computing systems fall short. With an IP67-rated enclosure, it is built to withstand extreme temperatures, vibration, and ingress from dust or moisture.
John Deere has also developed an air-cooled design for the unit that eliminates the need for liquid cooling or internal fans to reduce cost, complexity and maintenance.
Multiple configurations
Originally developed for use in John Deere’s See & Spray technology to distinguish in real-time between crops and weeds during spraying, the system is a scalable platform. It operates in conjunction with boom-mounted cameras and machine learning to enable precise, selective spraying, helping to reduce input costs and improve operational efficiency.
The VPU’s powerful 1 TB solid state NVMe drive and Nvidia core processor provide real-time image and video processing. It is equipped with Linux OS, 12 camera connections with power over coax (PoC), 4 CAN ports, 2 digital inputs, and 3 digital outputs.
OEMs are able to select the right balance of compute power and cost. The VPU is available in multiple configurations, ranging from high-performance systems capable of processing and switching quickly between multiple cameras, to more compact versions suited to simpler applications.
