Conferences and supporting programme
Centralized Raw-Data Sensor Fusion: A new Approach for a Safer Autonomous Vehicle Future
Automated driving (AD) is one of the most hyped technologies since the advent of the Internet. Buzz and anticipation around the technology seem to be reaching the “Peak of Inflated Expectations,” as evidenced by a June 2017 study estimating that AD technology will spur new industries and opportunities valued at $7 trillion by 20501. Hyperbole and hype aside, the promise of L5 automated driving is indeed profound. But there’s a problem that threatens to significantly delay and disrupt the era of autonomy: conventional sensing system approaches common to most AD platforms simply won’t work. These systems cannot fully support autonomous driving because they are based on a flaw inherent in the proposed sensing systems of virtually every commercially introduced AD platform. This is a very real problem regardless of whether the nodes utilize Lidar, radar, vision, or any other sensing technology. In these conventional sensing approaches, all sensor data is captured and pre-processed at the edge of the car’s data network (within each of the vehicle’s many onboard sensors). The data is then transported to a central unit for fusion and further processing. This presents two immediate problems: (1) the edge-processing at each sensor node discards large portions of the total data captured, resulting in insufficient data from which to establish the kind of detail sensory awareness upon which life-or-death driving policy decisions are made; and (2) the microcontrollers embedded in the sensors come with significant power and cost penalties. And since a self-driving car will have 20 or more sensors, these penalties are significant when compounded. After describing the design and inherent flaws to conventional AD sensing systems, this session presents an alternative AD sensor systems approach employing innovative “raw data sensors” which are unburdened by the power, cost, and size penalties of microcontrollers and related processing requirements in the sensor nodes. Eliminating pre-processing microcontrollers from all system sensor nodes enables a broad array of advantages including real-time performance, significant reductions in system cost and complexity, and access to all captured sensor data for the highest resolution possible of the vehicle’s environment and driving conditions. 1 https://www.theverge.com/2017/6/1/15725516/intel-7-trillion-dollar-self-driving-autonomous-cars.
--- Date: 28.02.2018 Time: 12:00 PM - 12:30 PM Location: Conference Counter NCC Ost