Robot localisation and sensor fusion
Uncertainty-aware odometry and visual, inertial and kinematic fusion for robots.
Role: Researcher, ARAS Labs, 2018 to 2022. I worked in a seven-person team spanning mechanical, control, electronics and mechatronics engineering. This includes my MSc thesis on cable-robot localisation.
Deep odometry with uncertainty
A Consistency-Based Loss for Deep Odometry Through Uncertainty Propagation (ICRA 2023) propagates pose uncertainty to balance incremental and trajectory-level losses. It improves pose accuracy and provides uncertainty estimates for localisation.
Localising cable-driven robots
- I fused position and orientation estimates from camera and inertial sensors with uncertainty-aware robot kinematics.
- With another engineer, I designed and implemented pose estimation for each sensor array. We also designed and built the robot’s end effector, data-acquisition system and joints.
- Graph-Based Visual-Kinematic Fusion and Monte Carlo Initialization for Fast-Deployable Cable-Driven Robots (IROS 2023) localises and calibrates cable-driven robots jointly, using factor graphs and Monte Carlo initialisation.
Other perception work
- I proposed an uncertainty-aware localisation subsystem for long-term and short-term odometry in autonomous-vehicle navigation.
- ARAS-IREF (ICRoM 2019, Best Paper Award) is an inexpensive, open-source infrared pose-reference system.
- ARC-Net (ICMLA 2020) fuses data from several inertial sensors with capsule networks for activity recognition. Code