Torc Robotics Forms DL+S Division to Unify Data, Simulation and ML for Autonomous Trucks

Torc Robotics created the DataLoop and Simulation (DL+S) division inside its AI organization to connect every link in the autonomous driving development cycle: data, simulation, machine learning, infrastructure and vehicle testing. As Torc progressed toward its AV 3.0 vision, it needed to unify data ingestion, processing, experimentation, training, validation and deployment across the organization, which required both organizational alignment and technical infrastructure able to orchestrate massive machine learning workloads. That led Torc to standardize on technologies such as Ray by Anyscale as part of its broader DataLoop strategy. The company says the approach required substantial upfront investment and cultural buy-in, but that standardized workflows already improve development velocity by enabling faster support and maintenance code, cutting duplication of effort. Common tooling also brings better auditability and repeatability, which Torc calls essential for safety-critical systems, and lets teams collaborate instead of rebuilding similar capabilities multiple times.

Simulation is embedded throughout several stages of the DataLoop, which is why it is called out separately in the division's name. DL+S uses simulation to support synthetic data generation, enabling teams to create validation scenarios that may be difficult, expensive or time-consuming to capture in the real world. Simulation is also the engine of reinforcement learning, allowing models to learn from large volumes of virtual interactions and gain hundreds of years of driving experience in hours. For model validation, replay and recompute workflows help developers analyze prior vehicle behavior, compare it against changes and test improvements under controlled conditions. Torc also lists the broader requirements: a data format rigid enough to scale yet flexible enough for future sensor and hardware changes, auto-labeling hand-checked by its manual labeling team, training frameworks that match the compute task to the right computing power so models train in hours rather than days, and simulation at all levels of fidelity for truck-specific motion models.

Many organizations focus on raw compute availability to scale workloads, but Torc argues the real challenge is using compute effectively: large multimodal training jobs must balance CPU resources, GPU resources and data throughput at the same time, and without efficient orchestration expensive infrastructure can sit underutilized while training waits for data or processing bottlenecks to clear. With Ray, Torc focuses on optimizing utilization rather than simply increasing capacity, shifting the question from whether it can get more hardware to whether it is using its hardware effectively. The company points to a recent optimization of a key workload for the scene modeling truck component, which provides behavior models with information about nearby objects such as vehicles and pedestrians. Felix Heide presented Torc's AV 3.0 at the 2026 Ray Summit, hosted by Anyscale.