DEV Community

Cover image for SLX SteerDrive: 3D EPS simulator and verification bench
Marco
Marco

Posted on Edited on Originally published at siliconlogix.it Fully Autonomous

SLX SteerDrive: 3D EPS simulator and verification bench

A virtual plant for the controller

Controller development needs controllable conditions: repeating a manoeuvre, changing a load, interrupting data and observing why assistance rises or becomes disabled. A steering animation shows movement; a software bench must also explain the relationship between input, measurement, firmware decision and plant response.

SLX SteerDrive is the Windows environment developed by Silicon LogiX to simulate electrically assisted steering and verify its control. It integrates an electromechanical model, 3D scene, repeatable scenarios, telemetry and recordings. It can use the software controller or the STM32H750B-DK controller, which actually computes assistance while exchanging data over Ethernet or CAN FD. The STM32 is the control hardware; motor, steering mechanism and load remain entirely simulated on the PC.

The simulator addresses practical questions: how much assistance does a given steering-wheel torque produce? How does it change with speed? What happens near an end stop or when a sample is missing? The simulated plant makes these conditions available without rebuilding a complete mechanical test rig for each experiment.

SteerDrive Windows application with a 3D steering bench, test controls and readings for torque, assistance, rack position and current

Frame from the recorded version 0.4.2 session in STM32 / PC model mode. The 3D view, torque, rack position and current describe the simulated plant connected to the controller. Application labels are in Italian.

From effort to motion

The model has two inertias, steering wheel and pinion, coupled through an elastic, damped torsion bar. Driver effort acts on the steering wheel; electric assistance is referred to the pinion. The sensor returns the elastic component of torsion, so during transients and reversals its reading can differ from driver-applied torque.

This avoids basing control on a measurement that simply copies the input command. The controller must use the sensor signal and system response. The rack converts pinion rotation into displacement; the geometric solver derives tie-rod, wheel and universal-joint poses from the same model state.

Steering model with steering-wheel inertia, torsion bar and sensor, pinion inertia, equivalent motor, rack and resistant load

The sensor measures torsion between the two inertias. The model separates controller-requested torque from the resulting electrical and mechanical response.

The load includes a speed-dependent restoring force, regularised friction and configurable resistance. End stops become progressive beyond 40 mm; the geometric domain ends at ±45 mm. These elements support analysis of assistance and reversals using estimated demonstrator parameters.

Virtual speed changes the assistance map and restoring load. The model does not simulate complete lateral vehicle dynamics: tyres, variable grip, slip and yaw require a further model and calibration. The project concerns EPS, electric power steering, and does not implement ESP stability control.

Motor, current and limits

The actuator is represented by an equivalent DC machine model with resistance, inductance, back EMF and current regulation. A transmission ratio refers speed and torque to the pinion, preserving power consistency between the shafts. The model exposes voltage, current and delivered assistance, separating the controller request from what the equivalent actuator can supply.

In the demonstration profile, simulated supply voltage is 24 V and nominal current limit is ±25 A. Electrical saturation and current response time influence model motion. The firmware receives simulated current and can reduce its command when it is high; fault scenarios also exercise supervision paths.

The model does not represent a three-phase inverter or FOC control. Current and torque cutoff at disarm is idealised; switching and magnetic-energy dissipation require a power-stage model. These boundaries define the results as control-bench tests, without suggesting validation on a real motor or vehicle.

Model, communication and graphics

The software separates model advancement, controller communication and rendering. Mechanics uses a fixed 1 ms step with implicit backward Euler integration; the PC worker advances ten steps per 10 ms update. The controller updates assistance at 100 Hz. Graphics frame rate does not determine the physics step.

The renderer reads a state snapshot and updates the scene and instruments. Communication checks that each command belongs to a recent sample and the current session. The model does not keep accepting the last command indefinitely when updates stop.

Software architecture with a fixed-step model, protocol adapter towards the controller, graphics and telemetry snapshots, and recordings for analysis

The model retains its numerical step. The MCU executes control and graphics presents state; Windows and networking remain part of the bench conditions.

A deterministic numerical step and real-time execution are different properties. Windows does not guarantee deterministic scheduling, so response times and cycles must be measured in the actual environment. SteerDrive is a bench with a hardware controller in the loop and a software plant, useful for functional verification and integration; it is not presented as a qualified hard real-time HIL system.

Reading the steering mechanism in 3D

The glTF/GLB scene makes the steering wheel, column, sensor, transmission, rack and wheels recognisable. Orbit, zoom and inspection follow the mechanical path. Live poses are calculated from model state; the renderer starts from neutral geometry to prevent accumulated transforms.

Tie-rod closure and universal-joint ratio belong to the verified kinematics. The view does not add decorative suspension deformation to suggest physics that is absent. Force and torque indicators relate movement to charted quantities.

Presentation mode includes six asset clips for exploring the project. Clips are separate from simulation and are not control evidence. This distinction keeps both bench architecture and live-session results readable.

Scenarios and manual commands

Eleven deterministic scenarios exercise the system with repeatable inputs. Operators can choose a test, set speed and load, compare Reference and Comfort regulation and observe changes and reversals. Repeating the same input helps relate a response difference to a regulation change.

Manual control includes torque, steering-wheel position and an axis reference. In Steering wheel mode a virtual driver follows the requested angle with limited torque. In Axis mode the controller regulates rack position while driver torque is zero. The model maintains one physical relationship between action and movement, without imposing both position and torque as independent mechanical constraints.

Fault scenarios introduce bench-declared conditions such as sensor anomalies or loss of the equivalent actuator. They exercise state, alarms, shutdown and reset. They do not constitute validated diagnostics for a physical sensor; real-device coverage depends on its interfaces and analysed failure modes.

Connecting firmware to the model

The SLX FD 1 protocol defines session identity, sequence, sample, transaction and values with explicit units. REQUEST → SENSOR → COMMAND relates the controller reference to the measurement and resulting command. HELLO/READY establishes peer availability; ARM and ACK confirm start, stop and reset actions.

On CAN FD, the payload is 64 bytes, with 11-bit identifiers and BRS at 500 kbit/s nominal and 2 Mbit/s data. The WinUSB/gs_usb backend handles the SH-C31G adapter; USB transmit-echo metadata is consumed locally rather than being interpreted as data received from the controller. PEAK PCAN is available as an alternative backend.

On Ethernet, the same application frame is encapsulated in an 80-byte UDP datagram. The decoder applies CRC, bounds, freshness and correlation checks on both paths. The firmware article describes frame format and IDs. CAN FD is a standard bus; SLX IDs and quantities are defined for this bench and are not automatically compatible with vehicle messages.

Ending a CAN session also requires closing the dialogue: the PC sends DETACH, waits for controller confirmation and releases the adapter. A subsequent connection creates a new session. Restored communication alone does not re-enable assistance.

Telemetry and diagnostics

Charts show driver torque, torsion sensor and assistance on a common time axis, alongside rack position, current and virtual speed. This helps identify saturated assistance, a reversal, an end stop or a request that does not produce the expected motion.

SteerDrive telemetry with separate torque, rack-travel, motor-current and virtual-speed charts

Actual application capture in software mode. Signals belong to the same simulated state; current retains the sign defined by the model's transmission ratio.

The Controller page distinguishes link availability from control enablement. Session, state, responses and counters help localise a problem: a detected USB adapter, an initialised CAN controller and valid application exchange are successive conditions to observe separately.

When a recent command is missing, the bench zeroes assistance and latches the fault. A new connection or duplicate packet does not erase the event. Explicit reset returns to the disarmed condition, from which the operator can choose to start again.

Recordings, comparison and replay

A session can be saved as a structured JSON recording, CSV time series, JSON report and HTML report with charts. Profile version, configuration and core identification accompany the results, so a reviewer can identify the conditions and version that produced the samples.

Replay displays recorded samples without rerunning the controller or enabling assistance. It supports reviewing a transient or discussing a fault. Exact sample reproduction is verified separately from running a new simulation.

Comparison applies the same inputs to the selected regulation profiles. Deterministic local comparison evaluates their responses; extending a conclusion to the physical controller requires the corresponding configuration and transport test, including timing.

One example is the Slow manoeuvre scenario, run for 10 s with version 0.4.2: the same input and model, with 10,000 steps per regulation profile. The following comparison is entirely software-based; currents, torques and displacements are model results with estimated parameters.

Simulated quantity Reference Comfort
Peak assistance 8.15 Nm 3.83 Nm
Equivalent motor peak current 4.94 A 2.32 A
RMS current over 10 s 3.48 A 1.63 A
Peak rack displacement 34.34 mm 18.38 mm

Comfort produces less assistance for the same driver input. Lower current accompanies different rack displacement; the quantities must be read together to assess the profile's behaviour. These are two responses to the same input, without treating an isolated current value as a measure of efficiency or overall quality.

The report associates results with each controller's configuration and the model version. A comparison can be reproduced after a change, separating a regulation change from a change in the simulated plant.

SteerDrive Recordings and replay page with recorded signals and controls for saving a session or opening a replay

Recordings preserve test evidence. Viewing a recording remains separate from commanding the plant.

How the software is verified

Verification covers contracts, numerics and operator workflows. Seven CTest groups check ABI, limits, transmission power, scenarios, command freshness, faults and replay. GUI checks cover nine pages, six clips and four About sections at three window sizes, plus actual pointer actions for start, pause, mode selection and recovery.

Kinematics was compared against an independent reference over 1,001 poses. For dynamics, two passive tests compare the implicit step with independent integration of the continuous equations: maximum rack deviation is below 0.004 mm. These are numerical-consistency checks, not accuracy measurements against a real steering mechanism.

Version 0.4.2 STM32–simulator verification recorded 5,959 valid COMMAND messages over CAN FD in 60 s, with a maximum observed response of 22.672 ms; Ethernet recorded 5,970 messages and 21.905 ms. On both links, an intentional 100 ms data interruption verified shutdown and subsequent explicit recovery. These measurements concern data exchange with the PC model, with motor and mechanics simulated. They describe the sessions performed and do not guarantee limits for every PC or network.

Tools for automotive and automation

The software uses a C++20 architecture with a version 1 C ABI SDK. Proprietary model and solver are encapsulated in the core, separate from interface, contracts and integration tools. This allows a demonstrator to be distributed without exposing model sources or STM32 firmware, while keeping versions and responsibilities identifiable.

Silicon LogiX develops simulators, test tools and technical applications around the client's device. SteerDrive's approach applies to ECU benches, axis controllers, electromechanical actuators and automation systems: define a repeatable plant, connect firmware, make signals observable and preserve test outcomes.

In automotive projects this can support embedded-function prototyping, integration and verification. Matching a vehicle requires manufacturer data, identified parameters and a measurement campaign. The demonstrator uses estimated parameters and does not claim ISO 26262/ASIL qualifications or AUTOSAR compliance. Its value lies in discussing behaviour using evidence and preparing product-specific verification.

Discuss your simulator or test tool

Technical references

Functions and results refer to SteerDrive 0.4.2 and verification performed on 12 November 2026.


Originally published on Silicon LogiX.

Top comments (0)