A modular, FMI 2.0-compliant validation framework that evaluates industrial robot trajectories against dynamics, safety, and smoothness criteria in real time within a ROS 2 / MoveIt 2 / Gazebo digital twin pipeline.
This repository contains the ROS 2 package fmu_validation accompanying the under-peer review IEEE Access paper:
S. Kahraman, C. S. Yilmaz, M. Yilmaz, and U. Yayan, "FMU-Based Multi-Criteria Trajectory Validation of Industrial Robots," IEEE Access, 2026.
The framework intercepts planned trajectories from MoveIt 2, resamples them to a fixed time step, and dispatches them to two parallel FMU-based validation modules before execution. This enables pre-execution assessment of trajectory quality, catching safety-critical deficiencies that are invisible to goal-position verification alone.
| Metric | Rapid Sprint | Chassis Inspection |
|---|---|---|
| PASS rate | 96.4% | 0% |
| Hard-fail ratio | 0.00 -- 0.40 | 0.53 -- 1.00 |
| Score paradox | --- | 96% soft-pass, 100% hard-fail |
The same planner and velocity configurations yield vastly different outcomes across scenarios, demonstrating the framework's discriminant power.
MoveIt 2 Planner ──► /display_planned_path ──► FMUValidatorNode
│
resample (Δt = 0.01 s)
│
┌────────────────┼────────────────┐
▼ ▼
DynamicsSmoothnessFMU SafetyFMU
┌──────────┬──────────┐ ┌────────────┬──────────┐
│ RNEA │ Jerk │ │ Joint │ Velocity │
│ Solver │ Analysis │ │ Margin │ Ratio │
│ (C++ FMU)│ (Python) │ │ │ │
└────┬─────┴────┬─────┘ └─────┬──────┴────┬─────┘
▼ ▼ ▼ ▼
S_dyn S_smo S_saf (two-tier)
Hard Gate + Soft Score
│ │
└───────────────────────┬──────────────────┘
▼
Weighted Score Aggregation
S = w_d·S_dyn + w_s·S_saf + w_m·S_smo
│
┌────────┼────────┐
▼ ▼ ▼
PASS WARNING FAIL
(≥0.70) (0.50-0.70) (<0.50)
│
IEC 61508-Inspired
Test-Level Aggregation
(Hard-Fail Ratio φ)
| Module | Method | Output |
|---|---|---|
| Dynamics FMU | C++ RNEA inverse dynamics solver (FMI 2.0 Co-Simulation) | Torque ratio score S_dyn |
| Safety FMU | Two-tier: hard gate at m_crit = 0.02 rad + geometric mean soft scoring |
Safety score S_saf |
| Smoothness | Length-normalized Gaussian jerk decay function | Smoothness score S_smo |
| Aggregation | AHP-weighted sum + IEC 61508 ratio-based hard-fail classification | PASS / WARNING / FAIL |
fmu_validation/
├── fmu_validation_core/
│ ├── __init__.py
│ ├── fmu_validator_node.py # Central orchestrator (ROS 2 lifecycle node)
│ ├── dynamics_smoothness_fmu.py # RNEA dynamics + jerk smoothness wrapper
│ ├── safety_fmu.py # Two-tier safety validation module
│ ├── fmu_backend.py # FMI 2.0 FMU loading and stepping
│ ├── common_utils.py # Shared trajectory utilities
│ └── validator_utils.py # Score computation helpers
├── scenarios/
│ ├── chasis_inspection.py # 18-waypoint chassis inspection runner
│ └── rapid_sprint.py # 4-waypoint rapid traversal runner
├── config/
│ ├── experiment_config.yaml # Full 450-test experiment matrix
│ ├── ur10e_params.yaml # UR10e joint specs and DH parameters
│ ├── validation_params.yaml # Thresholds and scoring parameters
│ └── weights/
│ ├── baseline.yaml # w_d=0.35, w_s=0.40, w_m=0.25
│ ├── safety_dominant.yaml # w_s=0.60 (HRC scenarios)
│ ├── dynamics_dominant.yaml # w_d=0.50 (high-payload)
│ ├── smoothness_focus.yaml # w_m=0.45 (precision assembly)
│ └── equal.yaml # w_d≈w_s≈w_m≈0.333
├── fmus/
│ └── ur10e_dynamics.fmu # Pre-compiled RNEA Co-Simulation FMU
├── msg/
│ ├── ValidationResult.msg # Per-trajectory validation output
│ └── ValidationMetrics.msg # Raw physical metrics
├── launch/
│ └── validation.launch.py
├── CMakeLists.txt
├── package.xml
├── setup.py
└── README.md
| Dependency | Version | Notes |
|---|---|---|
| Ubuntu | 22.04 LTS | Tested platform |
| ROS 2 | Humble Hawksbill | ros-humble-desktop |
| MoveIt 2 | 2.5.5+ | ros-humble-moveit |
| Gazebo | Fortress (Ignition) | Included with ROS 2 Humble |
| Python | 3.10+ | With numpy, scipy, pyyaml, pandas |
| FMPy | 0.3.x | pip install fmpy |
| UR packages | --- | ros-humble-ur-robot-driver, ros-humble-ur-description |
# 1. Create workspace
mkdir -p ~/colcon_ws/src && cd ~/colcon_ws/src
# 2. Clone repository
git clone https://github.com/ESOGU-SRLAB/FMU-Based-Trajectory-Validation.git fmu_validation
# 3. Install dependencies
cd ~/colcon_ws
rosdep install --from-paths src --ignore-src -r -y
pip install fmpy numpy scipy pandas pyyaml
# 4. Build
colcon build --packages-select fmu_validation
source install/setup.bash# Terminal 1: Start UR10e simulation + MoveIt 2
ros2 launch ur_simulation ur10e_sim_moveit.launch.py
# Terminal 2: Start FMU validator node
ros2 run fmu_validation fmu_validator# Chassis Inspection (18 waypoints, 225 tests)
ros2 run fmu_validation chasis_inspection
# Rapid Sprint (4 waypoints, 225 tests)
ros2 run fmu_validation rapid_sprint# Validation results
ros2 topic echo /trajectory_validation/result
# Raw metrics
ros2 topic echo /trajectory_validation/metrics
# Lightweight status
ros2 topic echo /trajectory_validation/statusWeight configurations can be changed at runtime:
ros2 param set /fmu_validator_node weight_dynamics 0.35
ros2 param set /fmu_validator_node weight_safety 0.40
ros2 param set /fmu_validator_node weight_smoothness 0.25Or load a predefined configuration:
ros2 param load /fmu_validator_node config/weights/safety_dominant.yamlThe full evaluation consists of 450 tests (2 scenarios x 225 each):
| Parameter | Values |
|---|---|
| Scenarios | Chassis Inspection (18 wp), Rapid Sprint (4 wp) |
| Planners | RRTConnect, RRT*, EST |
| Velocity scales | 0.075x, 0.15x, 0.30x |
| Repetitions | 5 per configuration |
| Weight configs | Baseline, Safety-dominant, Dynamics-dominant, Smoothness-focus, Equal |
| Level | Threshold | Condition |
|---|---|---|
| PASS | S >= 0.70 | No hard failures |
| WARNING | 0.50 <= S < 0.70 | Borderline quality |
| FAIL | S < 0.50 | Below acceptable quality |
| HARD FAIL | --- | m < 0.02 rad OR rho >= 1.0 |
| Regime | Hard-Fail Ratio (phi) | Decision |
|---|---|---|
| Normal | phi = 0 | Pure soft scoring |
| Random | 0 < phi < 0.50 | Dual condition: phi < 0.50 AND S >= 0.70 |
| Systematic | phi >= 0.50 | Unconditional FAIL |
Physical experiments on a real UR10e robot confirmed the FMU dynamics fidelity:
| Joint | Gravity Offset (Nm) | Comp. RMSE (Nm) | NRMSE (%) | Correlation r |
|---|---|---|---|---|
| J1 (Shoulder Pan) | 0.14 | 2.21 | 0.67 | 0.118 |
| J2 (Shoulder Lift) | 70.47 | 37.12 | 11.25 | 0.913 |
| J3 (Elbow) | 21.95 | 10.72 | 7.14 | 0.853 |
| J4 (Wrist 1) | 0.39 | 1.21 | 2.16 | 0.357 |
| J5 (Wrist 2) | -0.08 | 0.49 | 0.87 | 0.071 |
| J6 (Wrist 3) | -0.08 | 0.61 | 1.10 | -0.043 |
19,651 samples/joint over 56 minutes. The gravity offset arises because the UR10e controller reports post-compensation effort, while the FMU RNEA computes full inverse dynamics including gravity.
If you use this framework in your research, please cite:
@article{kahraman2026fmu,
title={{FMU}-Based Multi-Criteria Trajectory Validation of Industrial Robots},
author={Kahraman, Serhat and Yilmaz, Cem Suha and Yilmaz, Metin and Yayan, Ugur},
description = {Under Peer Review}
journal={IEEE Access},
year={2026},
publisher={IEEE},
}This project is supported by the European HORIZON-CHIPS-JU research project MATISSE and its members, including top-up funding by Vinnova (SE), FFG (AT), Business Finland (FI), Ministry of Universities and Research (IT), FCT (PT), and TUBITAK (TR).
This project is licensed under the BSD 3-Clause License. See LICENSE for details.
- Serhat Kahraman (Corresponding Author) -- serhatkahraman183@gmail.com
- Cem Suha Yilmaz -- cshyilmaz@gmail.com
- Metin Yilmaz -- metinyilmaz26@gmail.com
- Ugur Yayan -- ugur.yayan@ogu.edu.tr
Department of Computer Engineering & Software Engineering, Eskisehir Osmangazi University, Turkey