The Wheego AI Platform uses a mixture of HADOOP big data for storage and a GPU based super computer to train Deep Learning AI from data recorded within vehicles. Currently, the AI Platform is being used to train algorithms to provide perception for Advanced Driver Alerting System features and Autonomous control of vehicles. Another aspect of the system is a annotation server for the purposes of annotating images for custom training of Deep Learning detection AI. Examples include pedestrian, truck, car, cyclist, sign and road lane markers to provide a 3D perception of the environment surrounding the vehicle using Deep Learning frameworks.
As part of focus on autonomous systems, we have created drive-by-wire control of the vehicle for autonomous vehicle development. The drive-by-wire solution set integrates the correct automotive grade CAN systems with sensor sets and engine control units to enable autonomous control for self-driving and ADAS systems. To fully control a vehicle autonomously, one must have access to throttle, braking, and steering and the underlying position data from the sensors for vehicle algorithmic control. Wheego has create a solution for control of vehicles for both autonomous data collection and control needs, making autonomous systems easier.
Wheego has created a system to collect data used for autonomous and ADAS Deep Learning algorithmic Artificial Intelligence. Pioneering an approach to collecting the data required to make autonomous and ADAS products, data is collected and stored in our platform using two approaches. A low cost, custom-made device for collecting data crowdsourced at large scale and high resolution sensor suite for smaller more resolute data make up the data sets collected for autonomous algorithm training. The low cost way of collecting data results in the breath of data required for training for an autonomous vehicle to operate in all conditions. The high resolution data comes from vehicles outfitted with high quality sensors such as LIDAR, cameras, radar, ultrasonic, IMU, and vehicle controls. This data is used to provide AI training data to show the exact mapping within images of the vehicle to the physical world.
Using drive-by-wire, vehicle control, AI platform, and autonomous training data, a kit is provided for aftermarket integration with vehicles. While the kit is still in development and requires OEM partnership for integration, the autonomous kit offers actuators, control integration, camera, radar, and modular Deep Learning algorithms for control of a vehicle autonomously. Modularity is key for upgrading and ease of testing for market specific vehicle needs. Self-driving algorithms are created using the Advanced AI Platform with testability and safety at the forefront.