Blue-White Robotics (BWR) incorporated into its next generation of its agricultural autonomous driving kit, planned to launch in the coming months, Hailo’s AI microprocessor, designed to accelerate embedded AI applications on edge devices – says BWR’s CTO and co-founder Aviram Shmueli in a conversation with Techtime. BWR’s kit is designed to transform agricultural vehicles such as tractors into self-driving machines.
“This is the first time for an accelerator microprocessor to be integrated within an agricultural tool. Integrating the AI component makes it possible for us to overcome major limitations”. Integrated in the vehicle’s main computer, combined with a CPU and a GPU, Hailo’s chip will assist in performing processing tasks related to the sensors’ data and decision making, based on BWR’s proprietary models. “We’ve developed a significant AI layer. In order to achieve the desired level of autonomy, we combine neural networks and machine-vision in a magnitude that requires high computational capability at real time”.
Hailo have developed its neural networks Hailo-8 processor in a unique architecture which features processing power of up to 26 tera-operations per second (TOPS), with a typical power consumption of only 2.5W. The chip is capable of analyzing the particular task, mapping the required resources for each processing layer and to allocate them as necessary. This way the device accelerates the processing task and reduces power consumption.
Shmueli: “Conventional processors produce significant amount of heat, and we need to remember that these agricultural tools operate in relatively hot locations. Hailo’s chip features an excellent ratio of computational power per watt, which reduces the load. In order to achieve maximal computation capabilities, processing tasks are divided between the central & graphic processors and the acceleration card”.
Blue-White Robotics was founded in 2017 by Ben Alfi, Yair Shachar and Aviram Shmueli. It has raised a total of $50 million, in a round led by Insight Partners and Entrée Capital.
The automation kit developed by the company includes a package of LIDAR sensors, optical cameras, management software for detection and avoidance of obstacles, a pressure-sensitive front bumper to prevent vegetation damages, and it also equipped with emergency mechanisms and remote control capabilities. Using Artificial Intelligence, the tractor “learns” the land and available routs. Incorporating the kit in the tractor makes it possible for the vehicle to perform, autonomously or remotely controlled, a variety of operation such as sprays, harvesting, disking, and seeding.
Shmueli explains that autonomous field driving might be more challenging than one may describe. “Agricultural field is constantly changing, during the day and during the year, not to mention that this is a hybrid environment alongside human beings. This is why it is impossible to base field driving on a predetermined mapping. In addition, just like military vehicles, agricultural tools operate in harsh terrain conditions such as vibrations, high temperatures and moisture, and this is why computation units must be ruggedized”.