Bavantha Udugama 0000-0002-1932-692X

Bavantha Udugama · Robotics engineer · Netherlands

SLAM. Vision.
Navigation.

I build systems that help robots understand where they are, map their surroundings, and decide where to go. My work connects learned perception and visual-inertial estimation with embedded deployment and navigation in the field.

Explore selected work View CV ↗Contact ↗

Available immediately · Robotics roles & research collaborations

MONOCULAR 3D MAPPING / LIVE OUTPUTRGB + IMU
ITC corridor loop · Scene graph construction shown at 10× real time
One camera. A map of surfaces, objects and rooms.

Selected work / Research to deployment

Perceive. Localise. Navigate.

All projects ↗
Mono-Hydra++ · ScanNet scene 0000 · Playback at 16× real time

01 / SLAM & spatial mapping

From a camera stream to a structured world.

I developed Mono-Hydra++ to connect learned depth, visual-inertial estimation and hierarchical 3D scene graphs, then integrated the mapping stack on a custom drone.

  • Visual-inertial estimation
  • Semantic mapping
  • C++ · ROS
Camera + IMUPerception, odometry and scene graphs on Jetson
Inside Mono-Hydra++
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.

02 / Computer vision

Learn geometry and semantics together.

I develop multi-task perception models and evaluate their predictions inside a running mapping system. I take inference from PyTorch to ONNX and TensorRT on Jetson.

  • Multi-task learning
  • PyTorch
  • ONNX · TensorRT
25.53 fpsM2H-MX-L perception · 256 × 192 · Orin NX 16GB
Explore learned perception
Service robots at the Exness stand in Dubai. Expert Hub Robotics.

03 / Navigation & field deployment

Navigation beyond the lab.

I built navigation software for service robots, integrated sensors and vendor SDKs, and led a three-engineer team through customer deployments and changing crowds.

  • Navigation
  • Sensor integration
  • Field support
30+ robots10+ enterprise clients
Explore navigation engineering

Autonomous exploration / ATLAS

Teach a robot where to go next.

I developed ATLAS to select exploration goals from a persistent belief graph of places, open ends and unfinished work. The learned policy ranks routes admitted by a geometric planner.

Simulation validated · Drone transfer underway

Fully trained and tested in simulation, with earlier iterations tested on the drone. The current policy is now being transferred to the aircraft.

Explore the ATLAS approach
ATLAS autonomous exploration · Simulation demonstration

Connected expertise

From the model to the robot.

I connect learned perception, spatial mapping and autonomous exploration, then engineer the software for the robot’s hardware.

  1. InputCamera + IMUMonocular visual–inertial sensing
  2. 01Deep learning perceptionMetric depth · semantic labels
  3. 02Spatial AI and SLAMVIO · mapping · scene graphs
  4. 03Autonomous explorationImitation + reinforcement learning
Selected goals → planning and control → new observations
Deploy and support real robots.
The perception and mapping stack runs onboard the drone. ATLAS is trained and tested in simulation, with current drone transfer underway and earlier iterations flight-tested.

Embedded robotics / Hardware integration

Built for the aircraft.

Onboard architecture ↗
The quadrotor from a three-quarter angle, showing the depth of the carbon frame, the propeller cage, and the battery and electronics stacked in the centre.

Models, estimation and mapping.
On the robot’s own compute.

I built and flight-tested the custom drone, integrating camera–IMU sensing, learned perception, odometry and scene graph construction on Jetson.

  • Jetson Orin NX
  • ONNX · TensorRT FP16
  • C++ · ROS
Trace a frame through the stack ↗
Watch the aircraft in flight
Flight footage · Take-off and indoor corridor pass

The aircraft carries its mapping compute onboard. The ITC demo above shows the separate scene graph recording.

Research contributions

Selected publications.

All publications & results ↗
  1. Mono-Hydra++: Real-Time Monocular Scene Graph Construction with Multi-Task Learning for 3D Indoor MappingUnder review at ISPRS Journal of Photogrammetry and Remote Sensing
  2. M2H-MX: Multi-Task Semantic and Geometric Perception for Real-Time Monocular 3D Scene Graph ConstructionAccepted, ICRA 2026 SRRA Workshop
  3. M2H: Multi-Task Learning with Efficient Window-Based Cross-Task Attention for Monocular Spatial PerceptionPublished, IEEE/RSJ IROS 2025

Engineering & leadership

Build systems.
Lead their delivery.

I led a three-engineer team in industry, designed robotics labs at Twente, and supervised three master’s thesis students.

Experience & education ↗
Bavantha Udugama, in a dark suit and tie against a plain warm grey background.

Based in the Netherlands · Available immediately

Let’s build the next robot.

SLAM, computer vision, navigation and embedded robotics.

Get in touch