Abstract
The planning of safe trajectories in critical traffic scenarios using model-based algorithms is a very computationally intensive task. Recently proposed algorithms, namely Hybrid Augmented CL-RRT, Hybrid Augmented CL-RRT+ and GATE-ARRT+, reduce the computation time for safe trajectory planning drastically using a combination of a deep learning algorithm 3D-ConvNet with a vehicle dynamic model. An efficient embedded implementation of these algorithms is required as the vehicle on-board micro-controller resources are limited. This work proposes methodologies for replacing the computationally intensive modules of these trajectory planning algorithms using different efficient machine learning and analytical methods. The required computational resources are measured by downloading and running the algorithms on various hardware platforms. The results show significant reduction in computational resources and the potential of proposed algorithms to run in real time. Also, alternative architectures for 3D-ConvNet are presented for further reduction of required computational resources.
Highlights
Active safety systems for collision avoidance have a huge potential for increasing the road-traffic safety and are necessary components of vehicles for autonomous driving
In the Hybrid Augmented CL-RRT (HARRT) algorithm, 3D-ConvNet is used only for assisting the Augmented CL-RRT (ARRT) algorithm, but the actual safe trajectory planning is done by model-based methods which are interpretable and so suitable for safety-critical applications
ALTERNATIVE ARCHITECTURES In HARRT+ algorithm, 3D-ConvNet is chosen for extracting features from the input M as it is a suitable model for extracting the required spatio-temporal features
Summary
Active safety systems for collision avoidance have a huge potential for increasing the road-traffic safety and are necessary components of vehicles for autonomous driving. The Hybrid Augmented CL-RRT (HARRT) [22] and the Hybrid Augmented CL-RRT+ (HARRT+) [23] and GATE-ARRT+ [24] are examples of hybrid machine learning algorithms for long term safe trajectory planning in complex, critical traffic scenarios which use 3D convolutional neural networks (3D-ConvNets) [25], a type of DNN, to assist RRT variants the Augmented CL-RRT (ARRT) [26] and the Augmented CL-RRT+ (ARRT+) [23] to reduce the computation time Throughout this work, matrices are denoted by upper case bold letters, and vectors are denoted by lower case bold letters
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call