London-based Wayve, the leader in autonomous driving that the Virgin Group invested in, is joining forces with Microsoft to take advantage of their supercomputing infrastructure to support AI-based models for autonomous vehicles on a global basis.
With its data-driven ‘learned’ approach, Wayve is creating autonomous vehicle (AV) systems that can extend their driving intelligence to new, previously unseen places. This capability has already attracted two of the UK’s largest grocery retailers – Asda and Ocado – to trial the technology as part of their last mile delivery operations in London.
Scalability
Wayve, a London-based company that pioneers AI technology for autonomous driving, has announced it is joining forces with Microsoft to implement supercomputing infrastructure that will accelerate the global deployment of AI models for self-driving vehicles. This collaboration brings together Wayve’s industry-leading expertise using deep neural networks and vast amounts of data to train models, combined with Microsoft’s engineering prowess in creating large-scale AI systems.
Other AI platforms require costly hardware, HD mapping and complex localisation systems to operate in different cities, but Wayve’s data-driven “learned” approach enables its system to generalize – meaning it can adapt without prior city-specific adaptations. This simplifies AV deployments, allows more customers to benefit from AV technology and reduces driving complexity when unfamiliar places.
The London-based startup claims its AV2.0 technology is tailored specifically for fleet operators, featuring a camera-first approach with embedded AI that continuously learns from petabyte-scale driving data provided by Wayve’s partner fleets. This approach allows the platform to rapidly adapt to new driving domains and improve performance more rapidly than traditional AV solutions that rely on complex hardware and vast amounts of external data.
Wayve’s models are built upon video collected in real-time from cameras, with radar data serving as an additional input. It then utilizes open-source machine learning framework PyTorch combined with Microsoft Azure Machine Learning to collect, process and train its driving models at lightning speed. This scalable capacity has allowed Wayve to experiment, innovate and iterate quickly in order to deliver a market-ready solution within months.
In addition to speeding up its machine learning processes, the company also wants to make their technology simpler to maintain and use by automating routines like updating the model’s training data. Doing this will guarantee that the system runs reliably, securely and efficiently.
Scalability is one of the primary obstacles AVs must overcome when operating in multiple cities. Companies like Waymo and Cruise have been criticized for requiring separate sensors and modules for each city they operate in, which adds cost and complexity to their vehicles and makes re-engineering when drivers want to take vacations elsewhere more complicated.
Safety
Safety is one of the primary challenges faced by autonomous vehicles. That’s why Wayve ai is working to create a holistically learned driver, keeping their cars secure on the road.
This solution requires access to an expansive data set, powerful computation resources and sophisticated formulas for the model. To meet these demands, Microsoft’s Azure cloud has been chosen as its provider; providing these solutions.
The company aims to build AV2.0, an advanced autonomous driving system that can quickly and securely adapt to new driving domains worldwide. To accomplish this goal, they are relying on machine learning models that can learn to drive from scratch in any city.
Wayve’s approach has enabled it to expand its AV2.0 system into cities with varying weather, lighting and driving conditions. Furthermore, by eliminating HD maps and rules-based control strategies from their system, they reduce both cost and operational complexity associated with traditional AV platforms.
One key element of Wayve’s strategy is to break its AI into several smaller components. This allows the software to be tested across a range of scenarios and helps determine if a vehicle is secure for road use.
Therefore, the company utilizes PyTorch, an open-source machine learning framework, to train its models and process millions of hours of data. This information is captured through camera sensors in real time as well as radar and other sensor data.
Wayve’s AV2.0 models are trained using the MILE algorithm, a machine learning model capable of imagining and visualizing diverse and plausible futures. This helps the model plan its next moves accordingly.
This technology is helping Wayve reach their goal of bringing autonomous vehicles to 100 cities around the world, enabling them to expand much more quickly than with traditional AV systems that rely on HD maps and rules-based control strategies.
The company is also working with regulators to create a public scenario database, which will enable it to test its vehicles on the road and guarantee they are secure. Furthermore, this ecosystem of shared test scenarios can be utilized by various organizations to verify and validate automated vehicle safety.
Reliability
AI is an integral element of autonomous driving and has the potential to make vehicles more dependable than human drivers. Unfortunately, the industry has yet to create a platform capable of operating consistently under all conditions.
Wayve is working to change this by developing cutting-edge algorithms that learn from real world situations. Over time, the software should improve its abilities until it can operate as well or better than a human driver in realistic conditions.
To accomplish this goal, the company utilizes a combination of reinforcement learning and imitation learning to train its model. Computers observe humans driving and then duplicate their actions in order to drive autonomously.
Wayve’s technology differs from traditional self-driving cars, which rely on costly Lidar sensors and HD maps for local testing. They claim their system can adapt faster to new cities, types of vehicles and use cases than existing systems due to its rule-based architecture. As a result, commercial deployments can scale much faster than with traditional systems, the company states.
The UK-based company, which plans to be the first to deploy autonomous vehicles in 100 cities, recently raised $258 million in equity funding. It has been supported by Eclipse Ventures, Balderton Capital and renowned technology leaders such as Sir Richard Branson.
With this latest funding round, Wayve intends to continue growing its team, develop a Level 4+ prototype for passenger vehicles and delivery vans, launch last-mile delivery trials with Ocado and Asda, and construct data infrastructure that will enhance its core autonomy platform at fleet scale. Furthermore, they aim to launch their AV2.0 technology – which incorporates camera-first sensing with deep learning that continuously learns from petabytes of driving data – which was pioneered by them.
This computer vision system, which learns from observing human driving through reinforcement learning, is designed to be more general than other systems. It focuses on recognizing common elements in a scene and building clusters of understanding around them instead of relying on labelled data that may be difficult to interpret.
Flexibility
Wayve, a London-based autonomous vehicle (AV) pioneer, has built an advanced deep learning platform that enables it to rapidly test and refine AV models for cities. It utilizes PyTorch – an open-source machine learning framework – as well as Microsoft Azure Machine Learning to collect, manage and process millions of hours of driving data annually – including images, GPS data, and other sensor data.
Wayve needed an advanced infrastructure that could handle one-day training models with trillions of parameters and exabyte-scale image data from real world driving and simulation. That is why they are joining forces with Microsoft to implement supercomputing technologies that will power global autonomy in the future.
Wayve’s partners enable it to access vast amounts of training data through data collection devices installed in manually driven fleets. Furthermore, Wayve utilizes synthetic data created in-house for edge cases – unexpected situations that may not be easily encountered in real life.
This method has allowed it to successfully train its AV software on multiple simulations that simulate all kinds of edge-cases, from collisions with pedestrians and other vehicles to running red lights and stray dogs on the street. Furthermore, Ocado and Asda, two of the UK’s largest grocery retailers, have recently joined forces in testing this technology on their last-mile delivery vans.
Alex Kendall, Wayve’s CEO, believes that end-to-end deep learning AI is the key to scalable autonomous driving. It doesn’t rely on prewritten rules which could make proving safety in court difficult. Instead, Wayve’s AI uses a joint perception prediction and motion planning model that transforms input from cameras, lidars and other sensors into outputs like steering, braking and acceleration – similar to how student drivers learn to transfer their understanding of the world around them into accident-free driving.
The artificial intelligence (ai) is then trained on a simulator before being released into the real world. This allows it to adapt more quickly to unexpected conditions like weather, lighting, roadworks and other factors that may compromise driving safety.
Discover the best Autonomous Vehicle topics in IoT Worlds, click here.
