Safe Real-World Autonomous Driving by Learning to Predict and Plan with a Mixture of Experts
Authors:
Stefano Pini,
Christian S. Perone,
Aayush Ahuja,
Ana Sofia Rufino Ferreira,
Moritz Niendorf,
Sergey Zagoruyko
Abstract:
The goal of autonomous vehicles is to navigate public roads safely and comfortably. To enforce safety, traditional planning approaches rely on handcrafted rules to generate trajectories. Machine learning-based systems, on the other hand, scale with data and are able to learn more complex behaviors. However, they often ignore that agents and self-driving vehicle trajectory distributions can be leve…
▽ More
The goal of autonomous vehicles is to navigate public roads safely and comfortably. To enforce safety, traditional planning approaches rely on handcrafted rules to generate trajectories. Machine learning-based systems, on the other hand, scale with data and are able to learn more complex behaviors. However, they often ignore that agents and self-driving vehicle trajectory distributions can be leveraged to improve safety. In this paper, we propose modeling a distribution over multiple future trajectories for both the self-driving vehicle and other road agents, using a unified neural network architecture for prediction and planning. During inference, we select the planning trajectory that minimizes a cost taking into account safety and the predicted probabilities. Our approach does not depend on any rule-based planners for trajectory generation or optimization, improves with more training data and is simple to implement. We extensively evaluate our method through a realistic simulator and show that the predicted trajectory distribution corresponds to different driving profiles. We also successfully deploy it on a self-driving vehicle on urban public roads, confirming that it drives safely without compromising comfort. The code for training and testing our model on a public prediction dataset and the video of the road test are available at https://woven.mobi/safepathnet
△ Less
Submitted 3 November, 2022;
originally announced November 2022.
Integrating Small Satellite Communication in an Autonomous Vehicle Network: A Case for Oceanography
Authors:
André G. C. Guerra,
António Sérgio Ferreira,
Maria Costa,
Diego Nodar-López,
Fernando Aguado Agelet
Abstract:
Small satellites and autonomous vehicles have greatly evolved in the last few decades. Hundreds of small satellites have been launched with increasing functionalities, in the last few years. Likewise, numerous autonomous vehicles have been built, with decreasing costs and form-factor payloads. Here we focus on combining these two multifaceted assets in an incremental way, with an ultimate goal of…
▽ More
Small satellites and autonomous vehicles have greatly evolved in the last few decades. Hundreds of small satellites have been launched with increasing functionalities, in the last few years. Likewise, numerous autonomous vehicles have been built, with decreasing costs and form-factor payloads. Here we focus on combining these two multifaceted assets in an incremental way, with an ultimate goal of alleviating the logistical expenses in remote oceanographic operations. The first goal is to create a highly reliable and constantly available communication link for a network of autonomous vehicles, taking advantage of the small satellite lower cost, with respect to conventional spacecraft, and its higher flexibility. We have developed a test platform as a proving ground for this network, by integrating a satellite software defined radio on an unmanned air vehicle, creating a system of systems, and several tests have been run successfully, over land. As soon as the satellite is fully operational, we will start to move towards a cooperative network of autonomous vehicles and small satellites, with application in maritime operations, both \textit{in-situ} and remote sensing.
△ Less
Submitted 17 April, 2018; v1 submitted 30 October, 2017;
originally announced October 2017.
