Skip to main content

Oxa to use NVIDIA for autonomous vehicle software

Newly-announced Cosmos Predict models enhance Oxa's own AV training tools
By David Arminas March 27, 2025 Read time: 2 mins
Hub-to-hub logistics is one application for the technology (© Audioundwerbung | Dreamstime.com)

Oxa, a provider of autonomous vehicle software for industrial and commercial fleets will use NVIDIA Cosmos to accelerate industrial mobility automation (IMA) - the automation of repetitive driving tasks by work vehicles. 

These include fixed-route shared passenger transportation, airport ground transportation (baggage, freight, passenger/crew), port and retail yard trailer/container shunting, asset monitoring, factory line parts logistics and hub-to-hub truck logistics.

The tasks are currently performed by around 400 million work vehicles globally and are typically completed on uniform routes which are ‘place-specific’ - a known location - making them prime candidates for automation.

Oxa is using NVIDIA’s Cosmos World Foundation Models (WFMs), including newly-announced Cosmos Predict models, to enhance its own training tools. The tools, which includes Oxa Sensor Expansion, sit within its development toolchain, Oxa Foundry. Cosmos WFMs generate photo-real virtual world states as videos from multimodal inputs such as text and images.

Through its collaboration with NVIDIA, Oxa is able to generate vast amounts of diverse and realistic synthetic data. This expedites the training and validation of its software while significantly accelerating the development and deployment of safe, reliable and efficient self-driving solutions.

Oxa’s end-to-end system development approach, powered by Oxa Foundry, uses a novel GenAI approach to train and assure its self-driving software, Oxa Driver, ‘hyperlocally’ on planned, uniform and repeatable routes, transforming it from a generalist into a specialist.

These GenAI techniques enable the creation of representative and targeted ‘syllabuses’ for teaching and assuring Oxa Driver, with minimal and cost-effective source data requirements.

“By collaborating with NVIDIA and harnessing its latest technologies, we are accelerating our ability to deliver safe, reliable and efficient autonomous solutions to customers today, addressing critical challenges such as driver shortages and productivity gaps,” said Gavin Jackson, chief executive of Oxa. 

“The use of Cosmos for synthetic data generation combined with our own technologies will be instrumental in achieving our goals and unlocking the $2 trillion industrial mobility automation market.”

For more information on companies in this article

Related Content

  • Ford targets fully autonomous vehicle in 2021
    August 17, 2016
    Ford has announced its intention to have a high-volume, fully autonomous vehicle in commercial operation in 2021. The new vehicle will be a Society of Automotive Engineers-rated level 4-capable vehicle without a steering wheel or gas and brake pedals. It is being specifically designed for commercial mobility services, such as ride sharing and ride hailing, and will be available in high volumes. SAE level 4 is one level below full automation and is defined as ‘mode-specific performance by an automated
  • Ford splashes out $29bn on EVs and AVs
    February 18, 2021
    Connected vehicle delivery will also increase as part of automaker's deal with Google
  • Utah Department of Transportation: How we’re using traffic analytics software
    February 4, 2025
    Our use of Iteris ClearGuide lets our traffic operations engineers interpret critical probe traffic data without the need for statisticians and software developers
  • Reducing congestion with Tomtom's historical traffic data
    December 5, 2012
    Historical traffic data provided by TomTom is being used by the local government in Spain’s Basque region to reduce road congestion at less cost. Old habits die hard. Photos from as far back as the 1930s show people counting cars by the roadside in order to provide congestion data to those running road networks. Today, such techniques are still used, albeit augmented by a range of automation technologies such as inductive loops, infra-red sensors and number plate recognition. Even with these advances, howe