This repository contains the final report and demonstration videos for our course project in Robot Perception (May 2025), carried out as part of our M.Tech. program. The project focuses on comparing two prominent SLAM frameworks—Kimera and RTAB-Map—for indoor mapping and semantic localization using Intel RealSense data.
We explore Metric-Semantic SLAM techniques to reconstruct a 3D map of the ground floor of the TCS-X building, using RGB-D input from an Intel RealSense camera. Our objectives:
- Generate accurate, semantically annotated maps
- Compare the performance of Kimera and RTAB-Map
- Evaluate localization accuracy using Absolute Trajectory Error (ATE)
- Perform object-level localization (e.g., couches, people) using segmentation
Perception_report_Part_1__Falak_and_Ganga_.pdf: Detailed technical report covering our methodology, results, and analysis.- For videos refer: https://indianinstituteofscience-my.sharepoint.com/:f:/g/personal/falakfatima_iisc_ac_in/Eh4j7PTNXO1InAwmaEWQ_WwBF3Fdr1QTpbL60Ipvv_jqUQ?e=wiZo6u
- Kimera: Provided accurate visual-inertial SLAM with dense semantic mesh reconstruction.
- RTAB-Map: Enabled real-time RGB-D SLAM with effective loop-closure detection and point cloud maps.
- Segmentation Pipeline: Used DeepLabV3 with ResNet-50 backbone to detect and localize indoor objects in real-time.
- Falak Fatima (24349)
- Ganga Nair B (25565)
Indian Institute of Science – M.Tech. in Robotics and Autonomous Systems
Robot Perception, Jan 2025
Instructor: [Dr. Bharadwaj Amrutur]
⚠️ This repository does not contain source code. It serves as a documentation and media archive for the completed perception project.