I recently left @Cruise after 2 amazing years.

In that time, we launched & scaled cars with no driver to tens of thousands of happy customers. I'm super proud to have worked at such a special place.

What's next after 7 years in self-driving cars & AI, and what did I learn? 👇 7 years ago, self-driving cars were heavily rules-based, with limited use of ML (mostly for object detection).

On a good day, with perfect conditions, this produced impressive demos. But, place the car in a dynamic, complex scenario, and it fell apart.

Rules were insufficient!

A mistake I often see: people using self-driving car R&D timelines to estimate how long it’ll take to scale self-driving.

Yes, it took awhile to remove the Safety Driver—with a few missed milestones along the way—but scaling is going way faster… @Cruise was founded in 2013, and it took 7 long years of ML and robotics R&D to unlock 1 fully driverless AV in SF.

But, in just the last 12 months we expanded to 125(!) concurrent AVs, scaled from 1 city to 3 in <90 days, and grew # of rides 4x in the last quarter.

AI is changing everything, and it's not just ChatGPT or self-driving cars.

You'd be surprised at just how broadly AI is being applied across tons of industries.

I’ve now invested in 50 AI startups, and here are some recent investments using AI in fascinating applications 👇 ☄️ @ForgeAstro is building robots to autonomously mine asteroids, returning precious materials to Earth.

Why? One asteroid can contain up to $10,000 quadrillion(!) of rare material.

This is enabled by computer vision, affordable rocket rides, and small sat maturation.

3 years ago, I started angel investing in machine learning and autonomous robot startups. I'm now 35 startups in, learning a lot, and having a blast.

The future is going to be incredible. Here are just a few of the startups making it happen 👇 🚗 @AppliedInt builds software tools for automakers and AV companies, significantly accelerating their autonomy efforts. Think world-class simulation, validation, and more.

They are helping save lives at large scale, and just raised $175M at $3.6B.

blog.applied.co/blog-post/seri…

Our self-driving technology hit a big milestone recently.

The speed at which computer vision, robotics, machine learning, sensors, and compute is advancing is mind-blowing.

The result? Ride after ride after ride looking like this one 😍 Or this one…

🚨 It’s that time again!

Let’s dive into the latest release of @Tesla FSD (full self-driving) with a rather eventful trip through midtown Sacramento.

Thread... 1. Before we get started, here’s a refresher on two important differentiators for FSD.

These differentiators will (eventually) help FSD scale, but they also make the self-driving problem significantly more complex.

In this thread, we’ll explore examples of this.

https://twitter.com/olivercameron/status/1319835517094014977

🚘 Buckle in: it’s time to talk about @Tesla FSD!

In this thread, I’ll share what I find most impressive and the challenges that remain before the “F” in “FSD” is true.

Let’s go! 1. @Tesla FSD is a polarizing topic in AV for a few reasons:

✅ No use of pre-recorded HD maps
✅ Perceiving the world with cameras (no lidar!)

Many think FSD is very different than most other self-driving technologies, but you can boil it down to these two reasons.

Given the pushback from the AV industry on measuring progress with miles per disengagement, let’s talk about alternatives.

Inside @voyage, we measure our progress with something we call the Driverless Readiness Score (DRS).

How does it work?

https://twitter.com/olivercameron/status/1232799079450591234

The DRS represents progress toward a set of key metrics and deliverables. When goal metrics and deliverables are hit, we are ready for driverless operations.

What’s important is that this bundle of metrics & deliverables are chosen _specifically_ for our targeted roadway.

2019 was a game-changing year for self-driving cars. Many thought we were decades away from fully driverless cars, but they are here, now ✌️

Follow along as I recap just a sample of the amazing work accomplished within the industry in 2019. 1️⃣ @argoai released the super useful Argoverse.

https://twitter.com/olivercameron/status/1145464649401266176?s=20

✌️ Two years ago, we started a self-driving car startup. We were a little insane!

But it's paid off. The @voyage team has made ridiculous progress on our path to driverless. I wanted to share that progress and our plans.

2020 is going to be wild.

📰 news.voyage.auto/our-journey-to… 1️⃣ One day, our technology will be able to drive a car on any roadway safely.

Instead of waiting for that day before building a business, we are focused on partnering with large communities where we can commercialize the state-of-the-art in self-driving technology today.

👩‍💻 @argoai's release of Argoverse is game-changing for self-driving car research.

🗺 290km of mapped roadway
👀 11,319 tracked objects with raw sensor data
🧠 327,790 sequences of interesting scenarios
🔌 A thoughtful API to interact with the data

Let's explore... 1️⃣ Why is Argoverse special?

Argoverse is the most comprehensive dataset release of its kind, enabling deep experimentation in the fields of both perception and prediction.

What makes it so comprehensive? It includes HD maps!

📺 New video! Watch our self-driving system handle lots of complex agent interactions.

The @voyage team is making miracles happen. Our tech metrics are growing like crazy MoM thanks to them.

All clips were recorded in the last 10 days and are handled fully autonomously. Here's a few real-time clips...

🦌 Let's start with pesky deer running into the road.

🚀 The amazing @voyage team (close to 50 now!) has spent two years developing our self-driving technology.

In short: it's super impressive, and we're able to handle the complexity and edge-cases of our first market: retirement communities.

See for yourself. 1️⃣ Prediction, planning and interaction with agents is a _huge_ challenge, arguably up there with perception. We've had a world-class team working on this problem for over a year, and the result is so good.

🍎 I’ve seen a lot of headlines and charts about how far behind ’s self-driving car is (1.1 miles per disengagement!)

Join me as I try to explain why this is incorrect, and why  (according to their report) is perhaps in third place, as ranked by this terrible metric.  took the weird step of reporting two metrics at different periods in 2018.

1️⃣ Raw disengagements (until June 2018)
2️⃣ Important disengagements (from July 2018 onward)

Any program that’s been around for some time just reports 2️⃣ in their CA DMV report. Never 1️⃣.

🧠 @Waymo published a paper on a model called ChauffeurNet, which attempts to drive a car using imitation learning.

ChauffeurNet aspires to more human-like driving, in contrast to traditional robotics approaches which drive like...well...a robot.

sites.google.com/view/waymo-lea… Typical end-to-end approaches (raw sensor input to output raw control commands) struggle when faced with traffic rules (stop signs, traffic lights, etc.). ChauffeurNet aims to make progress with that problem to produce a more human-like driver that can handle complex scenarios.

📺 I attended my first @khoslaventures summit this year, and the quality of the speakers blew me away. TIL that many talks (from this year and past years) are published online!

These videos are an amazing resource for every founder out there. Here's a few to get you started... 1️⃣ This 2014 fireside chat with @vkhosla and @BillGates has only 26k views. That's nuts. There's so much knowledge transferred here. Go watch it now!