Bavantha Udugama 0000-0002-1932-692X

Research & engineering

Projects.
Map, navigate & deploy.

I build perception, mapping and navigation systems for robots, from the first model to deployment. These projects span my PhD research, commercial robotics and the work that got me started.

01 / Thesis researchUniversity of Twente · 2022–2026

Structuring the Seen, Exploring the Unseen

Monocular perception, 3D mapping and learned exploration, brought together on a custom drone.

The scene graph forming over one loop of the ITC second floor, from the first frame to loop closure. Playback is 10x real time. Live RGB, predicted semantics, and predicted depth run down the right.Watch the full clip

01.1 — Spatial understanding

Mono-Hydra++

2026 manuscript · Under review

I built a monocular scene-graph mapping system that combines learned perception with visual-inertial estimation. Sparse predicted-depth factors anchor the pose estimate, while predictions are aligned across frames before fusion. The ROS 2 stack runs on the drone’s Jetson Orin NX with a ZED X One camera.

  • C++
  • ROS 2
  • Factor graphs
  • Jetson Orin NX
Project & demos Code Manuscript
An indoor living room: a person seated on a patterned sofa, a table lamp, a wooden side table, an armchair, and framed pictures on the wall.RGB input
Predicted depth for the same living room, as a smooth colour field where the near armchair and sofa stand out from the far wall.Predicted depth
Predicted semantic labels for the same living room, with flat colour regions for sofa, chair, lamp, table, floor, wall, and pictures.Predicted semantics
Indoor frame, NYUDv2. Metric depth and semantic labels from one RGB image. Pick a layer, or drag to blend between them.

01.2 — Multi-task perception

M2H / M2H-MX

IROS 2025 · ICRA 2026 workshop

I designed how depth, semantics, surface normals and edges share information in one network. M2H uses DINOv2 with windowed cross-task attention; M2H-MX adds LoRA-adapted DINOv3 features and register-gated Mamba context. They provide the perception front end for Mono-Hydra++.

  • PyTorch
  • DINOv2 / DINOv3
  • Multi-task learning
  • TensorRT
M2H project M2H-MX project M2H code M2H-MX code
Autonomous exploration in simulation. The robot holds an occupancy map it has built itself and commits to one certified route at a time, so the map fills in region by region rather than by wandering. 60 second excerpt.

01.3 — Belief-graph based indoor exploration

ATLAS

Trained and tested in simulation · Current drone transfer underway

I developed ATLAS to choose exploration goals using a persistent belief graph. Trained through imitation and reinforcement learning, the current policy is fully tested in simulation and transferring to the drone. Earlier iterations were tested on the aircraft.

  • Reinforcement learning
  • Belief graphs
  • Isaac Sim / Isaac Lab
  • ROS 2
Exploration overview & video

The earlier Mono-Hydra system established the monocular mapping pipeline in 2023. Mono-Hydra project.

02 / Commercial roboticsExpert Hub Robotics · 2020–2022

Humanoid visitor guidance & interaction

I built the navigation software and led deployments of more than 30 robots for over 10 enterprise clients, including Expo 2020 Dubai.

Service humanoids running autonomously on the Exness stand in Dubai, navigating a floor full of people and furniture. Deployment work at Expert Hub Robotics, where I built the navigation stack and led the on-site rollout.

02.1 — Navigation, sensing & interaction

Robots working with the public

2020–2021 deployments

I integrated ROS and GMapping navigation, LiDAR and 3D obstacle sensing with robot SDKs and voice assistance. As the founding robotics engineer and later a three-engineer team lead, I worked on site to get the robots navigating and interacting reliably around visitors.

  • ROS
  • GMapping
  • LiDAR
  • Robot SDKs
Navigation case study Deployment videos & my role
The venue map and robot pose, built with GMapping SLAM.

Deployment coverage

Exness deployment · LeapRate, 2020 Expo 2020 Dubai · Gulf News, 2021

03 / Undergraduate researchUniversity of Peradeniya · 2017

Autonomous reconnaissance robot

My first exploration and mapping project, built in a three-student team.

The 2017 Peradeniya project video. A holonomic robot with a laser range finder and stereo cameras explores a plywood maze it has never seen, builds the occupancy map in the inset, and finishes with the stitched 3D reconstruction of the arena. The localisation and control segment, which was mine, starts at 0:59.Watch the full clip

03.1 — Bachelor’s research project

Mapping an unknown environment

2017

I developed the SLAM system, combining 2D LiDAR mapping with stereo vision for object geometry. The project produced three papers and received the IESL Best Undergraduate Project and IET Sri Lanka Best Technical Paper awards.

  • SLAM
  • 2D LiDAR
  • Stereo vision
  • Exploration planning
Robot demo Exploration planning paper Stereo vision & laser ranging paper