Issue 58 Uncrewed Systems Technology Oct/Nov 2024 WeRide Robotics | Simulation and testing | Orthodrone Pivot | Eurosatory report | WAVE J-1 | Space vehicles | GCSs | Maritime Robotics USV | Commercial UAV Expo | Zero USV

34 algorithms during failure modes, most often to drive the vehicle to a safe, nearby location. For development ease, Linux and QNX are used as the operating systems. Above this, situated in the Robobus’ automotive software between the vehicle’s shared memory layer and functional nodes, is Mariana, WeRide’s proprietary middleware platform. “At first, we used ROS, but there were a number of disadvantages that motivated us to develop our own middleware for smoother communications between the nodes and the shared memory,” Liu recounts. “For one, L4 autonomous driving requires us to handle massive amounts of sensor data, so we needed higher interprocess communications performance than ROS could achieve. We also wanted a decentralised architecture, so that if one functional node in the automotive software stack fails it won’t affect another node; ROS has one root master, so if the master fails, everything fails. “Mariana works without a master node: every node is functionally both a client and a master node, so there is no single point of failure and we’ve designed a lockfree structure into Mariana.” As a final point, controlling the middleware allows WeRide to better control the quality and hence consistency of data, which has been indispensable for accurately replaying (and hence analysing) batches of sensor data and subsystem telemetry in sequence. This helps improve both the self-driving algorithm, the internal networks and the company’s simulation systems. Across the network, a dual-redundant SAE J1939 CAN bus is used for comms between the MU and motor controller, while auto-grade gigabit Ethernet is used for the Lidars and cameras. “Ensuring an optimal network architecture took a lot of early planning, especially for the different levels and sections of the network, and for each of the specific kinds of channels each part would need to use – as well as rigorous and exhaustive tests for validation, obviously,” Liu says. “Signal integrity analysis tools were vital for seeing that signals were holding together well, and tools like Wireshark were really important for measuring the latency between different parts. Bandwidth management and signal prioritisation were largely handled manually, with just commonsense things like assigning vehicle-control signals a higher priority than sensor data being key there – if you’ve already detected a pedestrian on the road ahead of you, we need the vehicle to stop first and sense afterwards.” Motor matters The central drive motor is a permanent magnet AC machine, chosen for fine speed regulation, as well as the low weight and high power density of such systems. It runs on a 400 V bus; 800 V being excessive for the range and speed requirements of Robobus’s transport routes. Its peak power output is 120 kW, and such is its regenerative efficiency that WeRide estimates around a 21% range increase over the base, energyderived range estimate. In addition to regenerative braking, the brake-by-wire primarily functions using a conventional (by modern standards) combination of integrated brake control, electronic stability control and electronic parking brake systems, meaning a tripleredundant braking system, or quadrupleredundant if one counts the e-motor. Both steering motors are also permanent magnet machines, although each one features a dual winding stator configuration for extra redundancy and hence safety. The dual CAN bus extending to the motor-control systems from the MU principally monitors motor performance data (both in the traction and steering motors), braking performance and wheelspeed sensors. This data is processed via WeRide’s onboard analytics in real time to gauge whether control outputs are being executed correctly. The powertrain is capable of centimetre-accurate control, which Liu says has been achieved by repeated trialling to identify when and where errors originate during operation (such as localisation, perception and so on). “For example, to prevent errors coming from the inertia and movements of the vehicle, we first need to establish a very accurate vehicle dynamics model to fully understand the Robobus’ behaviours in different working conditions, and hence ensure that Robobus’ control strategy is actually in line with its physical characteristics and the effects that driving commands will have on how it’s moving,” he explains. October/November 2024 | Uncrewed Systems Technology Dossier | WeRide Robobus The brake-by-wire functions, using a combination of integrated brake control, electronic stability control, electronic parking brake systems and the e-motor’s own regenerative braking

RkJQdWJsaXNoZXIy MjI2Mzk4