The recent developments in the research area highlight a significant shift towards integrating artificial intelligence (AI) and machine learning (ML) techniques with physical and kinematic models to enhance the design, prediction, and navigation capabilities in various domains such as urban air mobility, autonomous driving, and robotic navigation. A common theme across these advancements is the emphasis on ensuring that AI/ML models adhere to physical constraints and real-world applicability, thereby improving their efficiency, accuracy, and feasibility. This integration not only accelerates the design and optimization processes but also enhances the safety, robustness, and interpretability of autonomous systems. Furthermore, the use of high-fidelity simulations and active learning approaches for training models without the need for extensive real-world data collection is gaining traction, showcasing the potential for significant reductions in computational resources and time.
Noteworthy papers include:
- A novel physics-constrained generative adversarial network (physicsGAN) for efficient and feasible takeoff trajectory design of eVTOL aircraft, significantly reducing computational time while maintaining high accuracy and feasibility.
- A hybrid machine learning model that combines deep learning with kinematic motion models for realistic and physically feasible trajectory prediction in autonomous driving, enhancing safety and robustness.
- A deep neural network approach for learning terrain traversability from simulation to real-world deployment, enabling effective robotic navigation in unstructured outdoor environments without the need for real data collection.
- An active learning enhanced surrogate modeling approach for the accelerated design of turbofan jet engines, demonstrating high accuracy with minimal relative error.
- A multi-objective ensemble-critic reinforcement learning method with hybrid parametrized action for autonomous driving, achieving compatibility across multiple driving objectives and enhancing overall performance and training efficiency.
