Robotics

Infrastructure-based perception

ActivePublished

Key innovation

Moves perception, prediction and planning off the vehicle into fixed environmental infrastructure, so the vehicle needs no onboard sensors or self-driving system.

How it works

Fixed sensors placed in the environment (e.g. LiDARs in a plant's infrastructure) continuously scan the space, localizing vehicles with centimeter-level accuracy and detecting and classifying objects. An offboard autonomy stack performs perception, prediction of object behavior (machine-learning models) and planning of safe trajectories. The computed trajectories are transmitted wirelessly to the vehicle at high frequency (10 times per second in Embotech's solution, in packets of about 150 bytes), where an onboard drive-by-wire control system executes them. A supervisory fleet management system assigns missions and orchestrates the movement of multiple vehicles.

Problem solved

Onboard autonomy requires an expensive sensor suite and compute in every vehicle and is limited by the vehicle's own field of view (blind spots). Shifting perception and intelligence into shared infrastructure allows autonomous movement even of unmodified vehicles, lowers per-vehicle cost and provides a broader, global view of the scene.

Components

Fixed infrastructure sensorsPerception and localization

Sensors (e.g. LiDARs, cameras) permanently mounted in the environment that observe the operating area, localize vehicles and detect objects.

Offboard autonomy stackPerception, prediction, planning

Off-vehicle compute performing perception, prediction and trajectory planning.

Vehicle-infrastructure communication linkControl-command transmission

Wireless link transmitting computed trajectories from the infrastructure to the vehicle in real time.

Fleet management systemFleet orchestration

Supervisory layer that assigns missions to vehicles and orchestrates traffic flow, integrated with the site backend.

Onboard actuation interface (drive-by-wire)In-vehicle motion execution

Minimal onboard system that executes received trajectories via electronic control of throttle, steering and brakes.

Implementation

Implementation pitfalls

Autonomy limited to the sensor-covered areaHigh

Because sensors are fixed in the infrastructure, autonomy works only within the instrumented zone; outside it the vehicle loses environmental perception.

Dependence on a reliable real-time control linkHigh

Trajectories are transmitted wirelessly to the vehicle, so link latency or loss must be handled by a safe fallback layer.

Evolution

2022

Infrastructure-based vehicle marshalling enters serial production

Inflection point

Embotech's Automated Vehicle Marshalling, built on an infrastructure-based autonomy model and TUV SUD-certified as Level 4, began operating 24/7 in serial production.