Innoviz is Educating the Future Workforce

Despite the size of the auto industry’s workforce, training a LiDAR how to drive is a hard task, and car experts with the right set of skills needed to do so are hard to come by.

Omer Keilaf, Innoviz co-founder and CEO, and Amir Day, Innoviz director for Computer Vision, presented to over 200 MIT students — from undergraduate to PhD candidates —to teach young engineers about building and training LiDAR for AVs. They spoke about sensor requirements, Innoviz’s perception software, which fuses computer vision with deep learning networks to enable safe Level 3 highway driving.

Wondering how to train a LiDAR? See below and watch Omer and Amir’s presentation on LiDAR hardware and deep learning requirements to learn more.

The above images demonstrate a classification algorithm for collision-relevant points. The LiDAR senses points in the distance that could cause a collision, and the vehicle knows to navigate around the point. This algorithm works well for static objects, but can’t stand alone to guarantee safety in highway driving scenarios.

The above images demonstrate a classic clustering algorithm, which estimates a 3D bounding box around any detected object. In the first image, it’s clear that the points indicate a pedestrian. A well-trained deep neural network would be able to tell that the second image is also a pedestrian. The third image could be anything. This is where we see that LiDAR is better than cameras, because the LiDAR doesn’t need to be able to classify the object, it just needs a couple of points to detect that it should avoid it.

The images above show the fusion of the clustering-based approach with deep learning detection-based approach. The blue box represents deep learning detection, but you can see here that it’s a few centimeters off. The white box represents the clustering detection algorithm and is tighter. Fused together, they create an accurate detection that includes classification.

You can go far with simple collision detection and clustering algorithms, but they’re still not enough to enable a safe automated driving experience, especially in high-speed, dynamic highway scenarios. Deep learning offers important information, including direction, speed and velocity, which allows the vehicle to perform everyday driving scenarios and be prepared for the unexpected.

AV technology is evolving and advancing rapidly, and it is important that younger generations be equipped with the right skills needed to maintain momentum for the industry. Innoviz regularly hosts high school students from around Israel that are learning about topics in STEM, such as AI and robotics, to expose them to new ideas and inspire them to change the world.

At Innoviz’s HQ, high school students are able to meet with experienced engineers to learn about potential career paths and see LiDAR technology often for the first time.