A Library of Lower Fidelity Dynamics Models (LFDMs) For On-Road Vehicle Dynamics Targeting Faster Than Real-Time Applications
Fuente:
Zenodo
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866902165679767552 |
|---|---|
| author | Unjhawala, Huzaifa Mustafa Mahajan, Ishaan Serban, Radu Negrut, Dan |
| author_facet | Unjhawala, Huzaifa Mustafa Mahajan, Ishaan Serban, Radu Negrut, Dan |
| contents | <p>A collection of low fidelity dynamic models, optimized for speed and efficiency, and primarily designed for robotics applications. It features models for wheeled robots, including vehicles, and is equipped with integrators (both half-implicit and implicit, utilizing Sundials) for simulating these models from a given initial state.</p> <p>Using CMake, the user can choose to build the models to execute on the CPU or NVIDIA GPU cards. The CPU models are implemented in C++, whereas the GPU models utilize CUDA. A Python API is also available, provided through SWIG-wrapped C++ models.</p> <p> </p> <h3>Key Features</h3> <ol> <li><strong><span>High-Speed </span></strong><strong>Performance</strong>: Models surpass real-time processing speeds. For instance, the 18 Degrees of Freedom (DOF) model achieves 2000x faster performance than real-time on standard CPUs, with an integration timestep of <code>1e-3</code> s.</li> <li><strong>GPU Optimization for Scalability</strong>: The GPU models are adept at parallel simulations of multiple vehicles. The 18 DOF GPU model, for example, can simulate 300,000 vehicles in real-time on an NVIDIA A100 GPU. Note: The GPU models are only available for Nvidia GPUs.</li> <li><strong>Python API</strong>: The SWIG-wrapped Python version maintains significant speed, being only 8 times slower than the C++ models, thereby offering Python's ease of use with C++ efficiency.</li> <li><strong>Advanced Analysis with Sundials</strong>: The CPU models support Forward Sensitivity Analysis (FSA) for select parameters. The use of a half-implicit integrator allows easy access to Jacobians of the system's RHS in relation to states and controls, beneficial for gradient-based Model Predictive Control (MPC) methods.</li> <li><strong>Comprehensive Vehicle Dynamics Simulation</strong>: Including models for the engine, powertrain, and torque converter, these simulations closely replicate actual vehicles. Users also have a choice between two semi-empirical TMeasy tire models, noted for their accuracy and performance at high vehicle speeds.</li> <li><strong>User-Friendly Configuration</strong>: Parameters for the models can be set dynamically at runtime through JSON files.</li> </ol> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_12703023 |
| institution | Zenodo |
| language | eng |
| publishDate | 2024 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | A Library of Lower Fidelity Dynamics Models (LFDMs) For On-Road Vehicle Dynamics Targeting Faster Than Real-Time Applications Unjhawala, Huzaifa Mustafa Mahajan, Ishaan Serban, Radu Negrut, Dan Reinforcement learning Vehicle engineering Controls <p>A collection of low fidelity dynamic models, optimized for speed and efficiency, and primarily designed for robotics applications. It features models for wheeled robots, including vehicles, and is equipped with integrators (both half-implicit and implicit, utilizing Sundials) for simulating these models from a given initial state.</p> <p>Using CMake, the user can choose to build the models to execute on the CPU or NVIDIA GPU cards. The CPU models are implemented in C++, whereas the GPU models utilize CUDA. A Python API is also available, provided through SWIG-wrapped C++ models.</p> <p> </p> <h3>Key Features</h3> <ol> <li><strong><span>High-Speed </span></strong><strong>Performance</strong>: Models surpass real-time processing speeds. For instance, the 18 Degrees of Freedom (DOF) model achieves 2000x faster performance than real-time on standard CPUs, with an integration timestep of <code>1e-3</code> s.</li> <li><strong>GPU Optimization for Scalability</strong>: The GPU models are adept at parallel simulations of multiple vehicles. The 18 DOF GPU model, for example, can simulate 300,000 vehicles in real-time on an NVIDIA A100 GPU. Note: The GPU models are only available for Nvidia GPUs.</li> <li><strong>Python API</strong>: The SWIG-wrapped Python version maintains significant speed, being only 8 times slower than the C++ models, thereby offering Python's ease of use with C++ efficiency.</li> <li><strong>Advanced Analysis with Sundials</strong>: The CPU models support Forward Sensitivity Analysis (FSA) for select parameters. The use of a half-implicit integrator allows easy access to Jacobians of the system's RHS in relation to states and controls, beneficial for gradient-based Model Predictive Control (MPC) methods.</li> <li><strong>Comprehensive Vehicle Dynamics Simulation</strong>: Including models for the engine, powertrain, and torque converter, these simulations closely replicate actual vehicles. Users also have a choice between two semi-empirical TMeasy tire models, noted for their accuracy and performance at high vehicle speeds.</li> <li><strong>User-Friendly Configuration</strong>: Parameters for the models can be set dynamically at runtime through JSON files.</li> </ol> |
| title | A Library of Lower Fidelity Dynamics Models (LFDMs) For On-Road Vehicle Dynamics Targeting Faster Than Real-Time Applications |
| topic | Reinforcement learning Vehicle engineering Controls |
| url | https://doi.org/10.5281/zenodo.12703023 |