As a machine learning engineer, Forrest Iandola has always cared about applying his skills to challenging problems, such as autonomous vehicles and climate change. He is also, admittedly, impatient.
In 2019, Iandola’s computer vision startup, DeepScale, was acquired by Tesla, where he joined the company’s autopilot driver-assistance team. But Iandola found himself wanting to make a more immediate impact.
In pursuit of higher velocity innovation, Iandola was drawn to defense tech company Anduril. Today he works as head of perception, training deep neural networks for Anduril’s products to understand their surroundings and provide operators with situational awareness. At Anduril his team can run daily counter-drone flight tests, identify problems and “try again as soon as tomorrow.”
“When I was at Tesla, we would release an updated neural network every month or two,” said Iandola. “At Anduril, we have been able to go from identifying a mistake to releasing a new neural network in as little as four hours.”
Iandola is focused on increasing defense efficiency through technology, which he believes can help pave the way for innovation across a multitude of industries. Built In LA checked in with Iandola to see what Anduril offers to engineers who are looking to increase the speed of technological advancement.
You have had a rich career in deep neural networks. What motivated you to join Anduril as the head of perception?
My previous role was at Tesla, where I worked on the autopilot driver-assistance system. I love working on challenging robotics problems, but I think full self-driving is a long way away. I am an impatient person, and I like making an immediate impact. I love that at Anduril we can go from idea to product in less than a year, and we are able to use machine learning to iteratively improve our products.
What does Anduril have to offer engineers interested in innovation?
A lot of companies in the defense industry tend to hamstring their engineers with slow processes and decision making. Engineers have to fight for access to enough GPUs to train neural networks. Some of these companies only run field tests occasionally. At Anduril, a big part of my job is making sure that things are progressing as quickly as possible, and I look for logjams and bottlenecks that I can clear to speed up development.
And it’s not just my team; the whole company moves fast. For example, on our counter-drone system, we have daily flight tests. We can evaluate the system, tune it and try again as soon as tomorrow. Sometimes the tuning is a quick change; other times we are solving tasks with machine learning that have never been done before, so it takes longer.
I look for logjams and bottlenecks that I can clear to speed up development.”
What is notable about working in the defense industry as an engineer?
A lot of our competitors are rooted in the hardware consulting business. Their tendency is to wait for the customer to tell them what to do, and then they build hardware to the customer’s specs. It’s really hard to build excellent products when you have a different product for each customer.
Our culture is quite different: We look for pain points that many customers have, and then we put our full weight behind developing a limited portfolio of products that can be sold to many customers. This leads to higher-quality products and a more focused engineering effort.
We put our full weight behind developing a limited portfolio of products that can be sold to many customers.”
What tools and resources are available for professional development at Anduril?
We give people as much responsibility as we think they can handle. It reminds me a bit of the movie The Aviator, where Howard Hughes hires a meteorology professor from UCLA to forecast and find the most clouds for filming a flight sequence for a Hollywood movie. But pretty soon the professor was helping to develop the Hughes Aircraft Company. I’ve seen a lot of people’s careers skyrocket like this at Anduril. They’re hired for their skills in a certain area, but they’re quickly given more responsibilities.
What is the impact of this work?
When choosing a company to join, one thing that was important to me is that we are building something real. I didn’t just want to join a company with a good story that has raised a lot of money; I wanted to join a company that builds things that customers love. I found that in Anduril.
When choosing a company to join, one thing that was important to me is that we are building something real.”
Describe a project you are particularly proud of at Anduril. Why is this important to you?
Deep neural nets work best when they are given the chance to run, make mistakes and learn from their mistakes. More concretely, we collect data from challenging scenarios, incorporate it into the training set and retrain the neural network. Operating with this level of speed means that we can be highly responsive to issues that arise in the field and it shows that our engineers are empowered to make rapid progress.
