Alex Kendall

Coordinating autonomous systems at scale is not easy. Ground robots in the Ukraine conflict are running into this problem. Rob Proctor's new Real-Time Innovations (RTI) blog does a great job explaining why & how to address it. It's an important read: https://lnkd.in/gdpTFS2w

27

1 Comment

Some of you know that we have been working on something big and we can finally talk about it. CommonAI has its first members: the UK's Advances Research and Invention Agency (ARIA) has joined the Scaling Inference Programme as part of a £50M project to bring down the cost of compute for AI and accelerate the route to market for selected startups. We are building a lab here in the UK to explore cutting edge hardware and software technologies and accelerate their route to market. The startups that we are working with in this programme will be announced soon so look out for more details, but in the meantime you can read about us in the Times or get in touch to find out more.

107

1 Comment

I am at the Institution of Mechanical Engineers (IMechE) in London at a conference on autonomous vehicles. The UK Government has allocated £135m of funding for autonomous vehicle trials and demonstration programmes - and they’re going to arrive in the UK far sooner than most people think. They are here, on London’s roads, right now. Currently in pre-launch testing, with Waymo and Wayve both collecting mapping and sensor data, and training the AI to recognise London-specific road features. We are set to see self driving taxis on the streets of London and Manchester, autonomous shuttle buses in Birmingham, Coventry and Milton Keynes and quarry vehicles in Scotland, all by this Spring. Over the past year, I have been heading up a team investigating what this means for heavy goods vehicles. With 63,000 job vacancies for HGV drivers today, and a further 150,000 HGV drivers due to retire in the next ten years, autonomous HGVs could provide huge benefits for this industry. What we have discovered has been far more profound. We're on the cusp of a technology that is so transformative, it will change the way we think of logistics, with significant cost savings and efficiency gains that could turn the industry upside down. #autonomousvehicles #ehgv #selfdrive VOLTEMPO Simon Smith Paul Jordan Jaynie Smith Neil Durno John Fox

97

7 Comments

💡 "If we knew what this was supposed to train for, we could do it better." That quote from a data annotator captures a critical challenge in developing AI perception systems for autonomous driving - and it's just one of many issues we uncovered in our latest research. I'm excited to share research I contributed to as an industry partner from Kognic, led by Hina Saeeda After 50+ hours of interviews across the supply chain - from OEMs to Tier-2 suppliers - we've identified 18 recurring data annotation errors that compromise system safety and reliability. Some errors were expected:  👉 Bounding box misalignments  👉 Wrong class labels  👉 Inter-annotator disagreement Others were surprises:  👉 Missing feedback loops (lack of close loop error handling, errors found but not corrected)  👉 Privacy/compliance omissions (GDPR requires face blurring, but this removes critical pedestrian appearance data needed for training robust detection models)  👉 Cross-modality misalignment (camera and lidars don't sync) What struck us most: these aren't just "data problems" - they're organizational, process, and tooling challenges that cascade through the entire AI lifecycle. Yet there's no systematic way to diagnose, track, or prevent them. Teams across the supply chain are dealing with annotation errors ad-hoc, without shared vocabulary or structured approaches. When we validated our findings with practitioners, the response was clear: they needed exactly this kind of framework. One OEM safety expert said it could be used for root-cause analysis of perception failures. A Tier-1 engineer called it a "checklist for setting up annotation pipelines." A Tier-2 director said it would improve client discussions and QA processes. That's the power of systematic approaches - they turn invisible problems into manageable engineering challenges. Proud to represent Kognic in this research - our commitment is clear: advancing autonomous driving requires not just better algorithms, but better data practices. Industry-academia collaboration like this is how we get there.  Thanks to Hina Saeeda, Mazen Mohamad and Eric Knauss for leading this important work, and to my colleagues at Kognic for supporting this industry-academia collaboration. #AutonomousDriving #AIEngineering #DataQuality #SafetyCriticalAI #ADAS

40

Waymo is going to London in '26. My experience in one in SF was amazing! Key takeaways: • Safety first: Waymo vehicles have 5× fewer injury-causing collisions and 12× fewer pedestrian incidents versus human drivers. • Local footprint: Engineering hubs in London and Oxford are already fine-tuning simulations for U.K. roads. • Strategic partnerships: Moove for fleet ops to Jaguar Land Rover’s all-electric I-PACEs • Global roll-out: After Tokyo tests in 2025, and U.S. services in Phoenix, SF, LA, Austin, there’s Miami, Dallas, Atlanta and more on deck. Why it matters to Infocusp Innovations: This aligns with what we’ve seen while building automotive AI solutions; you need to test on closed loop simulations with real-time data pipelines. Our experience in LLM-powered incident management and RAG-driven sensor analytics taught us that modular, secure cloud architectures can accelerate both compliance and innovation. Best Practices for anyone in AI/transport tech: 1. Invest early in local testing to adapt to unique traffic patterns. 2. Forge OEM and regulator partnerships to smooth the path to market. 3. Prioritize data-driven safety metrics to build public trust. How do you see autonomous ride-hailing reshaping urban life in the next five years? Read more: https://lnkd.in/gJn7Urc2

35

2 Comments

The shift from software-defined vehicles to AI-native architectures requires rethinking both software operational models and hardware computing constraints. This week I'm looking forward to joining The Gasgoo 盖世汽车 AI-Defined Vehicle Forum to discuss these challenges across two sessions: 🏢 The 4th AI-Defined Vehicle Forum 2026 📅 September 17, 2026 📍 Shanghai, China 🔗 Event details: https://lnkd.in/g_xz3hxT 🎤 Presentation | 11:30–11:50 CST "One Intelligence, Whole Lifecycle: The Missing Half of the AI-Native Vehicle"  Deploying AI inside the vehicle is only half the equation. To build a true AI-native architecture, we need the missing half: a unified cloud-edge infrastructure that continuously feeds, monitors, and optimizes intelligence across its entire lifecycle. 🎙️ Panel Discussion | 16:50–17:30 CST "The 'Impossible Trinity' of Automotive AI Computing - The Ultimate Game of Performance, Cost, and Power Consumption"  In the afternoon, I’ll join a panel of Chinese OEM leaders and industry experts to tackle a core engineering trade-off in centralized computing: balancing computing power with strict power budgets and cost targets. If you'll be in Shanghai for this event, I look forward to connecting with you. #Gasgoo #autotech #sdv #AIdv #vehicleAI

29

TONY HO, the founder of Osmosis AI, is featured in Mirror Review. Osmosis AI is a London-based company focused on retrofitting existing vans and trucks rather than building entirely new autonomous vehicles. Osmosis AI installs its Driver Agent on commercial vehicles at a significantly lower cost than purchasing new autonomous vehicles. This retrofit model makes autonomous driving more affordable and practical for logistics companies. Tony emphasises a Robotics-as-a-Service (RaaS) business model, in which fleet operators pay a monthly fee rather than a large upfront cost. Osmosis AI plans to begin automation in controlled environments, such as depot yards, before expanding to longer routes. Tony’s team includes Leo Francis, who leads electronics and PCB design and ensures vehicle robustness, and Mouneesh S., who designs the actuators and mounts that control the vehicles. Osmosis AI is progressing toward advanced AI models, such as Vision-Language-Action (VLA), for real-world driving. Tony’s long-term vision extends beyond autonomous driving to broader robotics applications, while remaining focused on delivering practical, incremental innovations that meet customer needs today. 🔗 Read the full story here: https://lnkd.in/eqiVgYqf

10

2 Comments