Einride is preparing to integrate an Nvidia artificial intelligence model into the system that drives its autonomous heavy goods vehicles, although it will not initially hand over control. On 21 October 2026, the Swedish truck manufacturer announced that it had begun evaluating Nvidia Alpamayo for use in Einride Driver, its autonomous driving system. Road testing is scheduled for the second half of 2027.
Alpamayo belongs to the vision-language-action (VLA) family of models, which interpret road scenes, propose the manoeuvres a vehicle should perform and explain the reasoning behind them. Within Einride Driver, the model will operate under the safety architecture developed by Einride, which retains full authority over the vehicle. Rather than replacing the existing driving system, it will therefore work alongside it as a component under evaluation. Einride will carry out further post-training and distillation of Alpamayo using its own operational data, adapting the model to its trucks and freight transport operating conditions. Tests will take place in shadow mode: the model will run in real time on the vehicle without issuing commands, while its proposed actions are compared with those of Einride's driving planner and with driver behaviour. Throughout this phase, control will remain with the existing safety systems.
"Our advantage has always been moving goods autonomously in the real world, for real customers," said Henrik Green, Einride's chief technology officer. He believes Alpamayo's reasoning capabilities could contribute to the development of Einride Driver and its safety architecture. Being the first company involved, he added, also provides an opportunity to help steer the model towards freight transport. The trials aim to shorten development cycles and reduce the cost of deploying Einride Driver across new vehicle types and routes, particularly motorway and point-to-point operations, within a network already serving customers.
The agreement extends a partnership announced on 21 September, when Einride selected Nvidia Drive Hyperion as the platform for developing the next generation of Einride Driver. Hyperion is Nvidia's reference computing and sensor architecture for Level 4 autonomous driving, which Einride will adapt for heavy goods vehicles. It incorporates the Halos safety system, while model training and validation involve Nvidia Cosmos, which Einride already uses to select camera data and generate synthetic scenarios, as well as Blackwell computing infrastructure accessed through an Nvidia-certified cloud provider. Einride remains responsible for developing the driving system, validating its safety and obtaining regulatory approvals.
Meanwhile, the Swedish manufacturer is expanding its autonomous vehicle trials. In Germany, it launched a regular service on public roads with Lidl in mid-September, using a cabless Level 4 autonomous electric truck with neither a driver nor a safety operator on board. The vehicle connects Lidl's distribution centre in Edermünde, near Kassel, with a local store, operating under a permit issued by the Kraftfahrt-Bundesamt (Federal Motor Transport Authority, KBA), the first of its kind granted in Germany. It can carry 15 Euro pallets and uses nine cameras, six radars and four lidar sensors for autonomous driving. During the planned four-month trial, it is expected to make up to three journeys a day, transporting more than 3,000 pallets of dry goods. The next step will be a multi-stop delivery route, while discussions are under way with the Schwarz Group, which owns Lidl, about extending the use of autonomous vehicles to other divisions.
P.R.
