No U-Turn Required: What Bidirectional Haulage Actually Changes

By: OpenAutonomy.com Editorial Team

A run of cabless, battery-electric autonomous haul trucks is arriving from Chinese OEMs and technology firms. The vehicles make the headlines. The more consequential shift is happening one layer down, in how autonomous fleets plan routes, draw geofences, and manage traffic.

The signal beneath the spectacle

Cabless, battery-electric, autonomous rigid and wide-body trucks have moved from concept to a recognizable product category. XCMG showed a cabless autonomous battery concept in 2023; WAYTOUS has demonstrated its diesel-free CarMo; SANY launched its battery-electric SKT145Ei in 2026; and, most recently, the AT-150 (a 136-tonne, bidirectional, cabless electric unit jointly developed by SPIC, Inner Mongolia North Hauler, and EACON) entered service at an Inner Mongolia open-pit coal mine. For a sense of scale, fully electric autonomous fleets are already in production, even if not yet bidirectional ones: a separate 100-truck electric fleet (XCMG trucks on Huawei's system) runs at the Yimin coal mine, so the electrification and autonomy this story rests on is no longer a trials-only proposition.

Strip away the “first in China” framing that accompanies most of these launches, and one capability is genuinely new rather than incremental: bidirectional operation, enabled by all-wheel steering, which lets a truck reverse its direction of travel without turning around. That single feature is less a vehicle story than a navigation-layer story, and that is where its consequences actually land.

What bidirectional does to the route graph

Autonomous fleets do not drive on maps; they drive on route graphs, networks of nodes and edges that encode where a vehicle may travel and in which direction. Conventional haulage assumes a truck has a front. Route graphs are therefore built with turning loops, switchbacks, and dedicated turnaround pads at loading and dumping points, and benches are widened to accommodate the swept path of a three-point turn or a loop.

A bidirectional truck collapses the turnaround. Segments that were effectively one-directional (with a turning structure at each end) can be modelled as traversable in both directions, and the turnaround geometry, along with the bench width it demands, can be reclaimed. In constrained corridors, in-pit ramps, and narrow benches where a turnaround simply does not fit, this is the difference between a route being drivable autonomously and not. The design question shifts from “where do trucks turn around” to “how do we sequence direction reversals safely.”

Geofences, swept paths, and a vehicle with no front

Removing the cab and making both ends symmetrical changes the geometry the geofence layer has to reason about. A vehicle with no privileged front has a symmetrical swept path and can approach a loading or dumping position from either end, so exclusion zones and human-exclusion buffers can no longer be biased toward a "forward" arc. They have to be symmetrical, and perception has to be genuinely 360-degree. The practical payoff is that the work a wider bench used to do can instead be done by tighter route graphs, geofences, and motion constraints. This is constraint-based navigation doing real operational work: the truck is not finding open space, it is threading a corridor defined by rules. It is also precisely the kind of narrow-corridor reality that early autonomous commissioning tends to run into first.

Meeting, passing, and the recovery question

Bidirectional trucks change traffic management too. With no fixed front, a truck's heading is something the system chooses, so when two meet in a single-lane corridor, either can reverse rather than yield. Direction becomes a variable to plan around. We've covered fleet-wide traffic coordination; the bidirectional case just adds that degree of freedom.

The cab removal has a quieter consequence worth naming. A conventional commissioning path moves a site from manual operation, to teleoperation-assisted running, to supervised autonomy, to full autonomy, and along the way, a cab is somewhere to put a person when something goes wrong. A cabless truck removes that rung. Exception handling and recovery become remote-teleoperation only, which raises the stakes on connectivity, on how the teleoperation station is designed, and on how gracefully the autonomy degrades when the link does. That is the same low-bandwidth-resilience question that keeps surfacing in incident reporting from remote and deep-pit sites, now made unavoidable by the form factor itself.

The case for open navigation schemas

Here is the throughline. As vehicle form factors diversify (directional and bidirectional, cabbed and cabless, diesel, hybrid, and battery-electric), the route-graph, geofence, and traffic-rule schemas become the layer that has to absorb the variety. If those schemas are proprietary to a single autonomous-haulage stack, then each new form factor increases the operator’s dependence on that stack, because the specialized navigation reasoning a bidirectional truck needs lives inside one vendor’s system.

The open-architecture argument gets stronger, not weaker, as form factors multiply. Standardized interfaces between fleet-management systems and autonomy platforms (a common way to express route graphs, geofences, and traffic rules) let a mine adopt a new vehicle type without re-platforming its navigation layer. Industry bodies are pointing the same way: the Global Mining Guidelines Group's implementation guidance treats interoperability as a key consideration for operators, and its newer mixed-fleet work (currently focused on underground) is organized around the coordination and safety layer shared across machines rather than around any single vehicle.

The honest caveat is that no published standard yet defines these navigation schemas the way ISO 23725 defines the FMS-to-AHS interface. The route graph, the geofence format, and the traffic-rule grammar remain largely vendor-specific today, so the open alternative this argues for is a direction the industry would have to build, not a shelf-ready standard an operator can demand tomorrow.

The trucks losing their cabs is the visible change, and it is a real one. The quieter, more durable question is whether the navigation layer they run on stays open enough to accommodate whatever form factor arrives next, because on current evidence, something will.