SBIR Phase I: Low-cost real-time perception system for self-driving consumer cars

该项目的更广泛的影响/商业潜力是在自动驾驶消费汽车中实际部署低成本、低功耗的实时感知系统。传感器中的这种边缘计算功能可实现真正自动驾驶所需的更高可靠性和更低成本的整体传感和计算。这种创新将大大有助于消费者尽早和广泛地获得安全和便利的好处。此外，先进的感知系统将对机器人技术产生潜在的长期影响，这可能导致创造新的市场和新的生活方式。这项小型企业创新研究（SBIR）第一阶段项目旨在开发实时感知系统的高效算法和软件实施，从而使自动驾驶汽车能够使用低成本的计算系统。该算法提供了一种利用图像特征进行同步定位和映射（SLAM）的新方法，计算成本比现有算法低100倍。它们还包括一个真正新颖的神经网络，用于融合图像特征和光探测和测距（LiDAR）特征，并执行目标检测，与最先进的方法相比，其复杂性降低了100倍。这些降低复杂度的算法可以在低功耗和低成本的SoC处理器上实现。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this project is the practical deployment of a low-cost and low-power real-time perception system in self-driving consumer cars. This edge computing functionality in sensors enables higher reliability and lower cost of overall sensing and computing needed for truly autonomous self-driving. Such innovation will contribute significantly to the early and widespread availability of safety and convenience benefits to consumers. Furthermore, the advanced perception system will have a potential long-term impact on robotics in general, which can lead to creation of new markets and new lifestyles.This Small Business Innovation Research (SBIR) Phase I project aims to develop efficient algorithms and software implementation of a real-time perception system to enable the use of low-cost computing systems for self-driving cars. The algorithms provide a novel way of using image features to perform simultaneous localization and mapping (SLAM) with 100 times less computational costs than the existing algorithms. They also include a truly novel neural network to fuse the image feature and light detection and ranging (LiDAR) features and perform object detection, which has 100 times less complexity compared to the state-of-the-art method. These reduced-complexity algorithms can be implemented on low-power and low-cost SoC processors.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.