Research group · Department of Computer Science and Engineering

Lynx

We build machines that have to work in places people already occupy — and the sensing, data and benchmarks that say whether they actually do.

Research areas Work with us
Host
BRAC University
Department
Computer Science & Engineering
Location
Dhaka, Bangladesh
Principal Investigator
Mollah Md Saif

01 The premise

A robot in an empty corridor is a solved problem. A robot in a corridor with forty students in it is not, because the space it is reasoning about will not hold still.

Lynx works at the point where sensing meets decision. A single sensor gives an incomplete and often wrong account of the world; several sensors disagree with each other. The interesting question is not which one to believe, but how to build a picture that stays usable while they argue.

That question shows up the same way in a rover crossing desert rock, a vehicle crossing an unmarked intersection, and a microphone array trying to pick one event out of a city's noise floor. We treat those as one problem with four faces.

02 Research areas

Robotics

Field and mobile platforms, and the unglamorous systems that keep them alive: locomotion and control, power budgets, manipulation, embedded compute, and hardware that survives being used outdoors rather than demonstrated indoors.

  • ROS
  • Control
  • Embedded
  • Field trials

Computer Vision

Detection, segmentation and 3D scene understanding under motion, crowding and bad light. We lift 2D foundation-model segmentation into 3D, label at scale, and keep the provenance and confidence of every label attached to it.

  • 3D detection
  • Segmentation
  • Auto-labelling
  • Calibration

Autonomous Driving

Ego-motion, multi-sensor fusion and perception for traffic that does not follow lanes. Most driving benchmarks come from a handful of lane-disciplined cities; we build data and evaluation for the streets where most of the world actually drives.

  • LiDAR–camera fusion
  • GNSS/INS
  • Datasets
  • Benchmarks

Audio & Acoustic Sensing

Sound as a sensing modality alongside LiDAR and cameras. Spectral analysis of the acoustic environment, source characterisation, and spectrum-based event detection for the cases where a scene is audible long before it is visible.

  • Spectral analysis
  • Event detection
  • Array processing

03 Current work

In progress · under peer review

Perception for dense, heterogeneous traffic

Autonomous-driving perception is measured almost entirely on lane-disciplined traffic from a few cities. Most of the world drives somewhere else — on unmarked carriageways shared by cars, three-wheelers, rickshaws, pushcarts and pedestrians, where the assumptions behind those benchmarks quietly stop holding.

We are building the instrument and the yardstick for that setting: a multimodal recording platform, an annotation pipeline that reports its own failure modes rather than hiding them, and an evaluation protocol that reports performance stratified by how crowded and how dark the scene is, instead of collapsing everything into a single number.

Full details, data and results are held back until the work clears peer review. We will link the release here.

  1. A

    Sensing platform

    360° LiDAR fused with stereo depth and multi-camera RGB, with post-processed GNSS/INS ego pose — built from parts whose limitations we measure and publish rather than assume away.

  2. B

    Annotation pipeline

    Largely autonomous labelling that lifts 2D foundation-model segmentation into 3D, and exits every stage clean, degraded with a named cause, or refused. Every released label carries its provenance.

  3. C

    Stratified benchmark

    Rather than assert that this traffic is harder, we measure it: metrics reported per density and illumination stratum, with a worst-case robustness score and zero-shot transfer from existing datasets.

04 Publications

Peer-reviewed work from the group and its members.

  1. 2025

    Ananna, N. S., Saif, Mollah Md, Noor, M., Awishi, I. T., Rhaman, Md. Khalilur, Alam, Md. Golam Rabiul

    Autonomous Navigation in Crowded Space Using Multi-Sensory Data Fusion

    2025 IEEE International Conference on Robotics and Automation (ICRA) · Atlanta, GA, USA · pp. 7815–7821

    A navigation stack for a vehicle moving through space that is busy with people: several sensing modalities are fused so that the picture the robot plans against stays usable when any single sensor becomes unreliable.

    DOI ↗

One submission is currently under review; it will be listed here once the venue's decision is public.

05 People

Mollah Md Saif

Principal Investigator · Adjunct Lecturer, Department of Computer Science and Engineering, BRAC University

Works on autonomous navigation and sensor fusion — machines that have to work out where they are and how to move through crowded places. Leads the group's driving perception, dataset and benchmarking work.

Students and collaborators

Lynx works with undergraduate and graduate researchers at BRAC University across robotics, perception and audio. If you want to build, instrument and evaluate systems that have to survive contact with the real world, get in touch — say what you have built before, and what you want to work on.

06 Contact

Email is the reliable channel. Research enquiries, collaboration and student applications all go to the same place.

Primary mollah.mdsaif@bracu.ac.bd Alternate mollah.md.saif@g.bracu.ac.bd
Address Department of Computer Science & Engineering
BRAC University, Kha 224 Bir Uttam Rafiqul Islam Ave
Merul Badda, Dhaka 1212, Bangladesh