Off-road Autonomy Requires Field Data
Scott Sanders, chief growth officer at Forterra, on the defense tech startup playbook
The core moat in off road defense autonomy is not model cleverness, it is hard won real world driving data tied to deterministic behavior. A ground vehicle has to handle mud, ruts, rocks, slopes, weather, and surfaces it has never seen before, then respond the same way every time so it can be tested and certified. That favors companies with years of field deployments, not companies relying mainly on generative AI to reason through rare cases.
-
On road autonomy benefits from giant repeatable datasets, lane lines, traffic rules, and millions of similar miles. Defense ground autonomy is a 3D off road problem, where the vehicle may be climbing a hill, crossing loose soil, or navigating in rain or snow with no prior map, which makes data collection much slower and edge cases much broader.
-
Forterra built around a single AutoDrive stack used across defense and industrial settings, which lets it collect operating data from battlefields, yards, ports, and private roads, then harden one repeatable system. The company says this same stack can be moved onto platforms quickly, which is the practical advantage of owning deployment data instead of just a model.
-
This is why adjacent defense autonomy leaders emphasize narrow, certifiable autonomy instead of open ended generation. Shield AI positions Hivemind around GPS denied, communications denied missions on specific aircraft and drones, where the job is to execute safely in bounded mission envelopes, not improvise like a chatbot.
The next wave of autonomy will use generative AI more as a tool around the stack, for simulation, labeling, software development, and operator interfaces, while the driving core stays tightly constrained and validated. As more ground systems enter production, the winners will be the companies that turn deployments into proprietary datasets and then into cheaper, faster certification and fielding.