Infrasturcture Enabled Autonomy (IEA)

Dec 30 2021: Abhishek Nayak


Abstract: IEA is a new paradigm in Autonomous Vehicle technology that looks at offloading the core computational capabilities of awareness generation and path planning from the vehicle out onto Smart Roadside Units equipped with various sensors. Through this distributed setup IEA provides a solution to shared liabilities by transferring the primary responsibility of localization from vehicle to infrastructure which in turn enables greater situational awareness of the area under the purview of IEA. IEA architecture is deployed on specific sections of roads or special traffic corridors by installing Road-Side Units (RSU) on either side of the road. These RSUs are fitted with multi-sensor smart packs (MSSP) containing sensors required for localizing vehicles. These MSSPs monitor the vehicles in the section of the roads under the purview of IEA and aid in generating situational awareness which can be transmitted to the vehicles subscribing to this information.

MSSP includes several sensors that carry-out specific individual tasks and aid in generating the localization information. For example, cameras installed on the RSUs as a part of the MSSP are used to monitor traffic by identifying and locating all the objects of interest in the traffic corridor. MSSPs are installed with special SmartConnect devices, whose function is to establish wireless connectivity between MSSPs and the vehicles subscribing to its information, thus enabling the transmission of information necessary for its localization. The SmartConnect devices are communication medium agnostic and modular so that they can be easily substituted by newer technologies.

In this project, I set up the Vehicle-to-Infrastructure (V2I) and Infrastructure-to-Infrastructure (I2I) communication networks using DSRC and developed the software stacks on smart infrastructures for vehicle state estimation, localization, and computer vision tasks like object detection, tracking, and semantic segmentation. Further, I calibrated the cameras and sensor systems using OpenCV and developed SLAM capabilities for RSUs by fusing IMU, camera, GPS/RTK, and odometry data using estimation filters (like Monte-Carlo and Extended Kalman Filters (EKF)) for autonomous navigation of a Lincoln MKZ vehicle.

References:

  1. Nayak, A., Chour, K., Marr, T., Ravipati, D., Dey, S., Gautam, A., ... & Rathinam, S. (2018). A distributed hybrid hardware-in-the-loop simulation framework for infrastructure enabled autonomy. arXiv preprint arXiv:1802.01787.
  2. Ravipati, D., Chour, K., Nayak, A., Marr, T., Dey, S., Gautam, A., ... & Swaminathan, G. (2019, October). Vision Based Localization for Infrastructure Enabled Autonomy. In 2019 IEEE Intelligent Transportation Systems Conference (ITSC) (pp. 1638-1643). IEEE.
  3. Gopalswamy, S., & Rathinam, S. (2018, June). Infrastructure enabled autonomy: A distributed intelligence architecture for autonomous vehicles. In 2018 IEEE Intelligent Vehicles Symposium (IV) (pp. 986-992). IEEE.