Home > Center for Intelligent Mobility (CIM) > Our Projects

Our Projects

  • CIM

Advancing Intelligent Transport

The Center for Intelligent Mobility (CIM) at KLE Tech is at the forefront of redefining transportation through innovation in autonomous systems, connectivity, and cybersecurity. Focused on building safer, smarter, and more sustainable mobility solutions, CIM blends AI-driven perception and control (ADAS/AD), secure communication frameworks, and real-world validation. From navigating complex traffic to building resilient vehicle networks, our work directly addresses the future of mobility. Industry partnerships with NVIDIA, LTTS, and BOSCH strengthen our impact, ensuring research translates into transformative real-world applications.
cim-advance-intelligence-banner
  • RESEARCH PROJECTS AT CIM

Vision in Action

CIM’s projects are built on a foundation of research excellence and future-focused innovation. We combine the power of AI, deep learning, and sensor fusion to address real-world mobility challenges—from making autonomous navigation safer to enabling precise spatial understanding. Our work is rooted in practical experimentation and collaboration, pushing the limits of what connected and intelligent transportation can achieve.
  • BEVFlow – Streamlined Excellence in Instance Prediction from a Bird’s-Eye View
  • BEVFlow tackles the challenge of motion prediction in autonomous vehicles using a unique architecture that integrates centripetal motion prediction with backward flow warping. The system uses a multi-view camera input to generate BEV (Bird’s-Eye View) feature maps, which are then processed to produce segmentation maps and future motion flows. A warping-based post-processing module helps reconstruct instance predictions with precision. With a strong focus on accuracy and efficiency, the project involves iterative optimization through hyperparameter tuning and training refinements.
  • Affordance Learning for Detection of Drivable Area
  • This project explores the detection of drivable regions using deep learning-based affordance learning. By training neural networks on multi-modal data from LiDAR, radar, and camera sources, the system maps low-level sensory inputs to high-level affordances such as free space, lane boundaries, and obstacle distances. By fusing spatial and temporal features using advanced convolutional and recurrent models, the system enhances real-time decision-making in dynamic driving environments. Robustness and generalization remain central to its development.
  • ReConfNET – Confidence-Driven SLAM for Scene Reconstruction
  • ReConfNET is a novel RGB-D SLAM framework developed for high-fidelity scene reconstruction in indoor environments. It utilizes a hierarchical generative model and a confidence-based loss function to refine pose estimation and depth prediction. With continuous RGB-D input, it dynamically selects high-confidence key points and uses neural implicit encoding with volume rendering to create scalable reconstructions. The result is a system that outperforms traditional SLAM methods in both accuracy and computational efficiency — with applications across robotics, AR, and intelligent environments.
  • RESEARCH FOCUS AREAS

Domains of Impact

01

Perception and Control

02

Security

03

Connectivity