Bryan Salesky

My Video Podcast episode on Software Defined Vehicles from the Ojo-Yoshida report is now available without a registration wall on YouTube. I talk about what I think SDV should mean, why SDVs are cool, software issues, architecture, and more. youtu.be/PmTFgZyPIEc

24

1 Comment

🚘🤖I found this presentation particularly insightful regarding the state of the art in autonomous driving. Mr. François Piednoël de Normandie outlined the methods currently employed to transition from Level 2 to Level 3 autonomous driving, and the significant challenges posed by progressing to Level 4. It may seem tempting to assume that a powerful AI algorithm could magically resolve all these issues. However, safety remains one of the most critical limiting factors in utilizing AI for autonomous driving. For these systems to be reliable, the behavior of the algorithms must not only be predictable but also verifiable. With this in mind, the contribution of formal methods to neural networks provides assurance of the algorithm's behavior within a specific domain. 🛫➕➖♾️ I am convinced that the certification of artificial intelligence algorithms will quickly become a major topic. Indeed, their potential contributions are incredible. However, their main limitation lies in their unpredictability. I believe the future will be shaped at the intersection of using algorithms across increasingly broader domains where they can be certified, alongside the use of more traditional methods, particularly those employed in the aeronautics field. It is obviously on this fascinating topic!

5

1 Comment

🚀 Big news for the autonomous vehicle industry! 🚀 Congratulations to Waymo on closing a monumental $5.6 billion funding round! This milestone is a powerful testament to the growing acceptance of autonomous technology in urban settings and to Waymo's dedication to making autonomous mobility both practical and safe. At Foretellix, we’re thrilled to see such strong investment in the AV ecosystem. Every commitment like this strengthens the industry and fuels a shared vision of a future where autonomy enhances our daily lives with reliability and trust. Waymo’s focus on scaling its robotaxi service responsibly—prioritizing safety and reliability—highlights the values that will create confidence in the future of autonomous transport worldwide. Looking forward to seeing Waymo grow in more cities and watching the entire industry evolve through these exciting times! 🌐 https://lnkd.in/eNm-TGzQ #AutonomousVehicles #AVIndustry #Waymo #Innovation #UrbanMobility #FutureOfTransport #SafetyFirst #Autonomy #Robotaxi #Foretellix

59

A study in perspectives on AV safety: a Waymo robotaxi approaches an intersection in San Francisco. Some other driver is doing "really fast donuts" in the intersection, and has been doing so long enough that the robotaxi rider looks up because of the smell of burning rubber entering the cabin. Robotaxi enters the intersection, then stops. The other driver halts and narrowly misses hitting the robotaxi (by perhaps a foot). The rider: "I can trust the Waymo, but I can't trust this driver". States that the other driver was upset but the robotaxi rider "shrugged, and pointed to the invisible driver." What ya' gonna do? It's a Waymo. "Still going to use Waymo." Presumed Waymo response to regulators and media: The human driver doing donuts was irresponsible. If there had been a crash it would have been the other driver's fault. (If I got this wrong would be happy to be corrected by Waymo. They have not commented as far as I know.) A broader safety engineering response: This is not the first time a robotaxi has encountered another car doing donuts in an intersection. In fact, a Cruise robotaxi was hit in a similar situation in March 2023 (SGO 30412-4960). From the description it seems there was plenty of warning this was happening. Why did the robotaxi not stop well clear of the intersection as a prudent human driver would have? Is Waymo monitoring SGO data and learning from what has been reported by other companies? Has Waymo included "other car doing donuts" in its hazard log? Did Waymo decide to ignore mitigation of this hazard for some reason? Will that change now that this has happened to Waymo instead of just Cruise? I'm glad in this case there was no injury, but there certainly could have been but for chance from a rare circumstance that is already known to have caused a crash. Being saved by the “grace of the Gods” is not a safety argument. I hope these questions are being asked and resolved by Waymo. But we won't know until the next donut incident. Primary Source: https://lnkd.in/eFHEMVFX Article: https://lnkd.in/eFVpfjec

65

14 Comments

Want to write log statements in single-digit nanoseconds? This cool paper from John Ousterhout's group at Stanford shows how: https://lnkd.in/gnkVXsTD I like it because it combines deep performance analysis with clever optimizations to make a system as fast as possible. NanoLog is implemented as a more-or-less drop-in replacement for printf that's ~100x faster. The key observation is that the vast majority of log messages are never actually read. Thus, NanoLog tries to do as little work as possible at logging time, deferring work to a postprocessor that runs when someone actually reads a log entry. When you log a message in NanoLog, it records only the dynamic part of the log message to an in-memory "staging buffer." To avoid overhead from synchronization or cache effects, these are single-consumer, single-producer per-thread circular buffers. Then, a background thread reads from each circular buffer, performs lightweight compression on each record, and writes it to disk. When you actually read a message, a postprocessor decompresses the records, interpolates them with the original static log message, and presents it to you just like a conventional logger. The paper has a detailed evaluation section showing which optimizations actually work. The key takeaway is that the combination of 1) only logging dynamic content, not static strings and 2) performing lightweight compression before writing to disk were the most effective optimizations. I wish more papers would do analyses like this, it's good science!

14

And just like that, Waymo announced they have an end-to-end model for autonomous driving. Former CEO John Krafcik states: 1. there is long-term promise for an e2e approach (e.g., interpretable decision rationale is possible), but 2. big challenges remain (insufficient long-term memory to solve complex long-tail challenges, sensor fusion), so 3. it's going to be super hard for e2e-only companies (Wayve, Waabi, Tesla) to catch Waymo, which has the only safely working AI approach in the space & a huge on-going advantage: Alphabet AI adjacency/talent

120

11 Comments

AI + Physics + Simulation It's what we're all about at Pasteur Labs! Check out Alexander's talk at CDFAM, where he explains how we improve engineering workflows, and how this accelerates R&D in critical industries like energy, national security, transportation, and more! Thanks CDFAM - Computational Design Symposium for the fantastic event!!

13

Tesla is one tech topic that has unconditional support. No matter how hard you point out some defects in it, people just love red cars. When I first began sharing my Tesla observations, people have been telling me that the info must be coming from Elon Musk haters. They might be right, as Tesla has been experiencing resentment after the hostile and rough takeover of Twitter. But, there are some simple truths that are hard to miss. Like pointing out the fact that Tesla experiences routine problems that have been solved by carmakers for decades now. Like the front wheel falling off, this and that common part not working as intended, etc. A dear friend shared with me that the Tesla he drove had water leaking from the roof when driving in the rain which left his documents wet. I mean, this is a solved problem. I am not encouraged to buy the bleeding edge of cars for my patience to be massacred by routine, casual, ordinary, dumb issues. Even people who encountered some Tesla issues are quick to defend Musk's cow. Maybe this issue that one other person reported was not exactly like that. Maybe this did not happen or it happened this other way. How did we get to this point? Maybe because they popularised novel features the right way? Maybe a lot of engineering went into it? Maybe the feeling is great when driving? Maybe. There are lots of reasons to love Tesla, but, when it's flawed, it's flawed. Love is blind. Red love blinds.

1

Tesla's Robotaxi Vehicle Will Address Expensive And Slow To Arrive Waymo And Uber Rides The Advantages of Tesla's New Robotaxi Vehicle: Revolutionizing Urban Mobility Elon Musk just stated the Waymo is "expensive" and "slow." He's not wrong... As we edge closer to a future dominated by autonomous vehicles, Tesla's anticipated robotaxi marks a significant leap towards redefining urban transportation. I'll explore how Tesla's approach to autonomous ride-sharing could reshape our understanding of cost efficiency, speed, privacy, and safety in transportation. Full Article: https://lnkd.in/gh7YMuzT $TSLA #tesla #evs #robotaxi #rideshare #tenten

4

A must-read for ML researchers and engineers. Robotics is a domain that keeps AI/ML research honest. If an AI/ML approach from your paper is applicable to robotics in principle but can't be deployed on a robot w/o "future work" 😉, keep on trying! This is also a reminder to self.

16

Thinking why Waymo takes another kind of approach. Do you think Waymo will produce its own vehicles? In the context of self-driving cars, Convolutional Neural Networks (CNNs) play a crucial role in perception tasks, particularly for processing visual data from cameras and other sensors. The CNN architecture enables self-driving systems to identify objects, interpret scenes, and make decisions based on visual inputs. CNNs are integral to self-driving cars, enabling them to process and interpret visual data from the environment. Their architecture supports object detection, scene understanding, semantic segmentation, depth estimation, and even end-to-end driving control, all of which are critical for safe and efficient autonomous driving. The real-time performance, accuracy, and ability to generalize from visual inputs make CNNs a cornerstone of modern autonomous vehicle perception systems. #selfdriving

2

https://lnkd.in/eewVd6Jw? Last week, a Waymo got stuck in front of a Vice Presidential motorcade and had to be manually driven away by a police officer. Which is mostly one of those stories that's only interesting because an AI did it rather than a person—it's definitely not the first time someone's been flummoxed by an unusual traffic pattern, and consequently had an emotional breakdown that made them incapable of getting out of the way. But it's also in a category of stories that will get more common over time: you'll hear a disproportionate amount about models behaving badly because 1) that will happen more often when it involves famous people, who are outliers in lots of ways, and 2) it's a bigger story because there's a famous person's name in the headline, even if they have nothing to do with it. Like many other models, Waymo's training tells it to memorize and then generalize. At first it's identifying special cases, and reacting to each of them, but over time it's finding connections between those cases—learning that the important thing about a car swerving into the lane in front of it is the swerving, not the make and model. The process of training a model that gently slows to a stop at a stop sign is more straightforward than training one that figures out how to anticipate how other cars will navigate around an accident or respond to an ambulance.

2

Using generative AI, NVIDIA operations built an AI Planner agent, developed on NVIDIA Inference Microservices (NIM). The agent leverages LLM, NeMo Retriever and CuOpt NIM to reduce re-planning time from hours to just seconds. https://nvda.ws/4bIfS2T

1

Using generative AI, NVIDIA operations built an AI Planner agent, developed on NVIDIA Inference Microservices (NIM). The agent leverages LLM, NeMo Retriever and CuOpt NIM to reduce re-planning time from hours to just seconds. https://nvda.ws/4bIfS2T

1

Changing out hardware shouldn’t be stressful. Imagine you’re using a motor controller that suddenly gets discontinued, or you want to upgrade to a different robotic arm. The last thing you should worry about is rewriting code or redesigning your system. At Viam, we ensure that our software remains hardware agnostic, allowing you to swap out components with ease, whether it’s motors, sensors, or entire robotic systems. This flexibility not only keeps your development on track but also future-proofs your product as new technologies emerge. You can adapt and scale your solution without being locked into a specific vendor or worrying about compatibility issues. We meet you where you are, so you can focus on building what’s next, not fixing what’s broken.

62

5 Comments