SBIR Phase I: Energy-Efficient Perception for Autonomous Road Vehicles

SBIR 第一阶段：自动驾驶道路车辆的节能感知

• 批准号: 1648576
• 负责人: Forrest Iandola
• 金额: $ 22.5万
• 依托单位: DeepScale, Inc
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-12-15 至 2017-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1648576&HistoricalAwards=false
• 关键词: SBIR; Phase; Energy; Efficient; Perception

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will be to allow consumers to buy vehicles enhanced by Advanced Driver Assistance Systems (ADAS) that are more robust and more accurate. Advanced Driver Assistance Systems in general, and fully autonomous vehicles in particular, promise a number of advantages, such as: (1) reducing the number of traffic fatalities in the US and abroad, (2) enabling humans to spend less time driving and more time on other activities, and (3) reducing fossil-fuel emissions. Practical implementations of Advanced Driver Assistance Systems require a few key elements: sensors, perception systems, motion planning systems, and control/actuation systems. Based on extensive discussions with key individuals at automakers and automotive suppliers, developing robust and accurate perception systems is the biggest obstacle toward developing mass-producible autonomous road vehicles.This Small Business Innovation Research (SBIR) Phase I project will create perception systems that utilize the rapidly evolving technologies of deep learning for computer vision. Specifically, the company will utilize deep learning to provide perceptual systems that are: 1) more robust in the presence of diverse and rapidly evolving sensor configurations; 2) more accurate due to the early fusion of sensor data; and 3) more accurate due to the application of state-of-the-art deep learning algorithms for computer vision. The company is already engaged in developing partnerships with automotive OEMs and semiconductor suppliers that will enable it to deliver proofs-of-concept of its unique approach.

这项小型企业创新研究（SBIR）I阶段项目的更广泛的影响/商业潜力将允许消费者通过高级驾驶员援助系统（ADAS）增强的车辆，这些车辆更强大，更准确。总体而言，尤其是完全自动驾驶汽车的先进驾驶员援助系统承诺有许多优势，例如：（1）减少美国和国外的交通死亡人数的数量，（2）使人类能够更少的时间在其他活动上花费更多的时间，以及（3）减少化石燃料的排放。高级驾驶员辅助系统的实际实现需要一些关键要素：传感器，感知系统，运动计划系统和控制/驱动系统。基于与汽车制造商和汽车供应商的关键人物进行广泛讨论，开发健壮而准确的感知系统是开发质量可实现的自动驾驶汽车的最大障碍。该小型企业创新研究（SBIR）I阶段项目将创建感知系统，它将利用对计算机视觉进行深入学习的快速发展的技术。具体来说，该公司将利用深度学习来提供：1）在存在多样化和快速发展的传感器配置的情况下更健壮； 2）由于传感器数据的早期融合而更准确； 3）由于将最先进的深度学习算法应用于计算机视觉，因此更准确。该公司已经从事与汽车OEM和半导体供应商建立合作伙伴关系，使其能够提供其独特方法的概念证明。

项目成果

SqueezeDet: Unified, Small, Low Power Fully Convolutional Neural Networks for Real-Time Object Detection for Autonomous Driving

• DOI: 10.1109/cvprw.2017.60
• 发表时间: 2017-01-01
• 期刊: 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW)
• 作者: Wu, Bichen;Iandola, Forrest;Keutzer, Kurt
• 通讯作者: Keutzer, Kurt