PhD in Robotic Engineering

Policy gradient methods, sim-to-real transfer, and curriculum learning for multi-fingered grasping and in-hand object re-orientation in unstructured environments.

Leveraging large vision-language models (LLMs + ViT) for open-vocabulary instruction following, scene understanding and long-horizon task decomposition in household robots.

Behaviour cloning, GAIL and diffusion policy approaches for teaching robots complex manipulation skills from a handful of human teleoperation demonstrations.

Latent-space world models (DreamerV3, RSSM) enabling sample-efficient policy optimisation from offline datasets without environment interaction during training.

Dense visual SLAM using neural radiance fields for photorealistic scene reconstruction and real-time camera localisation in indoor and outdoor environments.

Distributed pose graph optimisation and inter-robot loop closure for swarms navigating underground mines, disaster sites and indoor warehouses.

Spatio-temporal occupancy grids fusing LiDAR point clouds and camera features via transformers for autonomous driving perception and motion forecasting.

High-resolution tactile sensor arrays combined with contact-rich reinforcement learning policies for precise fingertip object control without visual occlusion.

UniAD and VAD-style architectures unifying perception, prediction and planning in a single transformer network trained on large-scale driving datasets.

Deploying quantised neural networks on Jetson Orin and FPGA for agile MAV obstacle avoidance and target tracking in GPS-denied, cluttered environments.

Chance-constrained trajectory optimisation and probabilistic risk maps for mobile robots navigating safely among unpredictable pedestrians in hospitals and offices.

Domain randomisation and adaptive teacher-student distillation for training quadruped and biped robots in simulation that transfer directly to real hardware without fine-tuning.

Surface EMG signal decoding with LSTM and transformer networks for voluntary motion intent detection in upper-limb rehabilitation exoskeletons for stroke survivors.

Force/torque sensor-guided tissue manipulation, visual servoing and learning-from-demonstration approaches for laparoscopic subtask automation with safety guarantees.

Graph neural network-based pedestrian trajectory prediction and social force models for proxemic compliance and natural path planning in shopping malls and airports.

EEG-based motor imagery decoding with FBCSP and EEGNet for controlling 6-DOF robotic arms, enabling paralysed users to perform ADL tasks with minimal training time.

Stigmergy-based and bio-inspired coordination algorithms for heterogeneous UAV-UGV swarms in search-and-rescue missions with intermittent communication.

FEM-based design optimisation and data-driven control of silicone-based soft robotic grippers and limbs that safely interact with fragile objects and human skin.

Privacy-preserving federated RL enabling distributed robot fleets in warehouses and hospitals to collectively improve shared manipulation and navigation policies.

Elastic weight consolidation and progressive neural network architectures for mobile robots that continuously acquire new skills without catastrophically forgetting old ones.