Realtime Robot Localization and Pose Regression with Invertible Neural Networks

Details

• Alternative Title:
  Project 564: Realtime Robot Localization and Pose Regression with Invertible Neural Networks
• Creators:
  Mangharam, Rahul
• Corporate Creators:
  University of Pennsylvania
• Corporate Contributors:
  United States. Department of Transportation. University Transportation Centers (UTC) Program ; United States. Department of Transportation. Office of the Assistant Secretary for Research and Technology ; United States. Department of Transportation. Federal Highway Administration
• Subject/TRT Terms:
  Autonomous Vehicles Data Quality Location Neural Networks Normalizing Flows Robotics Robot Localization
• Publication/ Report Number:
  564
• Resource Type:
  Tech Report
• Geographical Coverage:
  United States
• Edition:
  Final Report (July 1, 2024 – June 30, 2025)
• Corporate Publisher:
  Carnegie Mellon University. Traffic21 Institute. Safety21 University Transportation Center (UTC)
• Abstract:
  This project explores novel approaches to robot localization and visual pose regression using Invertible Neural Networks (INNs). Addressing the critical need for efficient and accurate pose estimation in robotics, we propose two frameworks: Local_INN and PoseINN. Local_INN tackles the inverse problem of robot localization by providing an implicit map representation in its forward path and performing localization in the inverse path. It uniquely offers uncertainty estimation through latent space sampling and addresses the kidnapping problem with a global localization algorithm. PoseINN extends this work to real-time visual-based pose regression from camera data. By leveraging INNs and normalizing flows, PoseINN achieves state-of-the-art performance with significantly reduced computational costs, enabling faster training with low-resolution synthetic data and real-time deployment on mobile robots. Both frameworks demonstrate that INNs can effectively solve ambiguous inverse problems in robotics, providing robust and efficient solutions with inherent uncertainty quantification.
• Format:
  PDF
• Funding:
  69A3552344811, 69A3552348316
• Collection(s):
  University Transportation Centers
• Main Document Checksum:
  urn:sha-512:30ba66b1867fca31c4a2b19417f316340c02d67a1626c110651394b1023a30a04ba6721aa510257a5c8b1bae3dc6914a3239932df7ff4bea988a96d7896543c2
• Download URL:
  https://rosap.ntl.bts.gov/view/dot/85745/dot_85745_DS1.pdf
• File Type:
  [PDF - 2.48 MB ]