comma.ai Blog

thru the footer i discovered https://blog.comma.ai/autonomy/ which i have to call out for clear simple communication which i loved

Congrats! Just read up on thr taco bell drive. Great achievment! https://blog.comma.ai/taco-bell/

I'm not sure whether it's quite in that category yet - or whether this quote demonstrates that it's an elaborate joke/troll related to Silicon Valley (the TV series), but progress on Comma Body[1] may be worth following. "One day we dream that 3 comma bodies will get into an openpilot car, drive to Phoenix, open a donut shop, run said donut shop, and bring the profits back once a week. The potential of AI. But today, it balances, rolls around, and looks beautiful." [1] - https://blog.comma.ai/commabody/

The linked [1] blog post is more structured and easier to follow for context. After that, I think you could parse much of geohot's stream of consciousness post. [1] https://blog.comma.ai/end-to-end-lateral-planning/

Chatgpt: # don’t own the cloud, rent instead the “build your own datacenter” story is fun (and comma’s setup is undeniably cool), but for most companies it’s a seductive trap: you’ll spend your rarest resource (engineer attention) on watts, humidity, failed disks, supply chains, and “why is this rack hot,” instead of on the product. comma can justify it because their workload is huge and steady, they’re willing to run non-redundant storage, and they’ve built custom GPU boxes and infra around a very specific ML pipeline. ([comma.ai blog][1]) ## 1) capex is a tax on flexibility a datacenter turns “compute” into a big up-front bet: hardware choices, networking choices, facility choices, and a depreciation schedule that does not care about your roadmap. cloud flips that: you pay for what you use, you can experiment cheaply, and you can stop spending the minute a strategy changes. the best feature of renting is that quitting is easy. ## 2) scaling isn’t a vibe, it’s a deadline real businesses don’t scale smoothly. they spike. they get surprise customers. they do one insane training run. they run a migration. owning means you either overbuild “just in case” (idle metal), or you underbuild and miss the moment. renting means you can burst, use spot/preemptible for the ugly parts, and keep steady stuff on reserved/committed discounts. ## 3) reliability is more than “it’s up most days” comma explicitly says they keep things simple and don’t need redundancy for ~99% uptime at…

the picture of that GPU installed on the side of the firewall tunnel reminded myself and my husband of a gadget they would have shown off in the original Batman series. "Golly gee whilikers Batman, what is this high tech contraption?" "It's a car tracker my good lad, see how sleek and easy to hide it is?" https://blog.comma.ai/img/chestnut/install_footwell.png

I’m naive to the 10x gain value benefit to self driving. I’m excited for it nevertheless. Less than $1k for better cruise control/lane centering. Now, less than $2k for LIKELY some self driving magic SOONISH. All-ish open source. https://comma.ai | https://blog.comma.ai/chestnut/

Very interesting article. I'm looking to get my first robot arm soon-ish, probably something in the SO-101 category. Can someone get reasonably far using recorded sessions compared to training in a simulated environment¹? Do you have any experience with third-party or DIY attachments for robot arms? I assume it's going to be more difficult for something like the Ufactory arm vs. the open models. ¹https://blog.comma.ai/mlsim/

Here's some context for why those park assist commands are so dangerous. https://blog.comma.ai/safer-control-of-steering/

Comma is actually hosting a hack in February, if anyone is feeling up to trying something like this on another "vintage" car https://blog.comma.ai/comma-hack-5/

Going to leave this here https://blog.comma.ai/a-100x-investment-part-1/ https://blog.comma.ai/a-100x-investment-part-2/

comma.ai | Software + Hardware Engineers | San Diego, CA | ONSITE Our mission is to solve self driving cars while delivering shippable intermediaries. We build and sell the comma 3X to a growing fleet of users and get back 3M minutes of driving data every week for training openpilot. Some background on what we're hiring for: https://blog.comma.ai/refactoring-for-growth/ https://comma.ai/jobs - Systems Software Engineer - Full Stack Software Engineer - Electronics / Hardware Engineer - Controls Engineer - Designer

https://blog.comma.ai/a-100x-investment-part-1/ https://blog.comma.ai/a-100x-investment-part-2/ It's basically a bunch of apples-to-oranges comparisons that don't make sense.

I don't think that's what Hotz is saying at all, but he's saying that he's not person for the job anymore. He writes: "My two part blog post from 2019 remains a true description of the space." And this is a quote from (part two of) that blog.[0] > Nobody will have cars without a safety driver in cities anytime soon. Self driving appears to be like fusion, but instead of being perpetually 50 years away, it’s perpetually 5 years away. It’s 5 years away because that’s what the marketing departments are told, but the truth is finally leaking out. It’s going to be decades before robotaxis are profitable, and these companies are burning way too hot to last decades. [0] https://blog.comma.ai/a-100x-investment-part-2/

I saw this in Comma AI's blog post - they have also been sued and are planning to fight back: https://blog.comma.ai/ourfirstlawsuit/

A lead would need to teach significantly more to even be considered. Does the vehicle have (1) an after-market device installed that (2) detects that a first component stopped sending commands to a second component and (3) uses relay switch to disconnect the failed first component from the second component and (4) starts sending commands to the second component on its own that are indistinguishable from those sent by the first component? Technically, the above requirements are needed against claim 20 of '707. Claim 6 requires fewer features. That doesn't help us because: (1) claim 20 is asserted in the complaint [0], and (2) claim 20 was upheld in a previous challenge to its validity (an "IPR"), so it's got some staying power. [0] https://blog.comma.ai/img/20220824085232055.pdf

https://blog.comma.ai/commabody/ The comma body is the first “intermediary” for the new mission. It’s also the pathway to supporting all cars that a human can drive. The endgame of comma isn’t a box that sits on your windshield, it’s a robot person that gets in the driver’s seat. And it will be able to do a lot more than drive. Appears to me to be a distracting pivot. Especially given the “Stay Focused on the Mission” statement in the part 1 article (which are worse examples, but still I think relevant).

Don't miss part 2: https://blog.comma.ai/a-100x-investment-part-2/

Apparently it's an... indoor self-driving car? For software testing?? > The comma body runs our open source software, openpilot. From openpilot’s perspective, the comma body is a car. It runs the same camera and logging software, so we’ll be able to learn from the fleet of bodies in the same way we learn from our fleet of cars. https://blog.comma.ai/commabody/

Apparently some just gamble that by having some instances of every time, they won't all go down: "However, your cloud provider can pull these virtual machines out from under you at any time without notice (because someone else is willing to pay more). So how do you lower the probability of losing all of your instances to nearly zero? Cloud providers generally need to have a large buffer of capacity and they have many different virtual machine configurations with different CPU and memory sizes. So, don’t just spin up virtual machines of a single configuration type, spin up lots of them from many different configuration types!" https://blog.comma.ai/scaling-for-10x-user-growth/ I'm not sure if they still have a fallback of traditional instances.

Relevant: https://blog.comma.ai/011release/

>Each robot had several human escorts, and the robots were limited to slow walking and a few slow hand gestures. More importantly, the robots were limited to doing no real work. They just feebly pick up objects and place them somewhere else, which I am pretty sure doesn't require AI. For example, the vid shows the robot pouring hot water into a glass with a massive funnel strapped to it. Why not have the robot fill the kettle, place the teabag itself, etc? It seems like the kind of thing that should be developed before walking and talking and telling jokes. What if the refrigerator, microwave, etc. could interface directly with the robot. For example, the refrigerator has some type of robotized shelf that is able to bring a rack of orange juice to the front before the robot comes over to grab it? What if the microwave is able to focus the microwave beam on the food to cook it evenly? It also irks me how the robots are just humanoids. Like for example, why have a head with two eyes. Does it need to wear a helmet? Does it need exactly 2 eyes at exactly human-like placement to achieve stereopsis? Why not have 3 eyes? Did the designers think about the form of the machine at all, or did they just produce robots in the form that is associated with the most hype and thus will bring in the most investor capital? Is this really the ideal form for interfacing with humans? With other robots? I am just very skeptical of these companies that want to go from zero to doing everything.…

COMMA | Software Engineers | ALL FULL-TIME | ONSITE IN SAN DIEGO ONLY comma is a profitable company solving self-driving cars and robotics. We make open source software called openpilot and manufacture the comma 3X hardware onsite in San Diego. Think Tesla Autopilot like functionality for your Toyota, Honda, and more. See our open source driving agent on GitHub: https://github.com/commaai/openpilot Watch it take us to Taco Bell on YouTube and read our blog post about it: https://www.youtube.com/watch?v=SUIZYzxtMQs&t=2s We're hiring great software engineers to: ---> Make a great web experience: comma.ai, connect.comma.ai, etc. ---> Run our factory: https://blog.comma.ai/comma3X/#reliability ---> Ship openpilot releases like this: https://blog.comma.ai/094release/ ---> Write good infrastructure: https://blog.comma.ai/dev-speed/ Our stack is largely Python, with some C/C++ and JS(React) Interview Process (roughly) = Coding challenge + 2 phone screens + paid micro-internship on site Visit https://comma.ai/jobs for more detailed job descriptions ---> Reach out with a GitHub/LinkedIn to [email protected] ---> Fast Track by submitting a solution to the Calibration Challenge: https://github.com/commaai/calib_challenge

Part 2: https://blog.comma.ai/a-100x-investment-part-2/

I saw a video of a Tesla swerving around a jaywalking pedestrian the other day too, relatively high speed, also scary It is interesting to compare the various systems safety pages - https://waymo.com/safety/ - https://www.tesla.com/safety - https://www.ford.com/technology/driver-assist-technology/ - https://www.gmc.com/connectivity-technology/super-cruise - https://www.chevrolet.com/support/vehicle/driving-safety/driver-assist… - https://github.com/ApolloAuto/apollo (Baidu based) - https://github.com/commaai/openpilot/blob/master/docs/SAFETY.md | https://blog.comma.ai/understanding-the-openpilot-safety-model/#:~:tex… The last two's minimalism around safety is not reassuring