Ranbao Deng

X-Frontiers · Netherlands

• Helping to bring an AI-driven rescue drone from prototype to field deployment (~16 h/week)

University of Notre Dame · Indiana, USA

• Simulated tractor performance on different soft soils with terramechanics models in Webots
• Supported AgJunction's wheel-navigation products with simulation results and a literature review

TU Delft, Cognitive Robotics · Delft, Netherlands

• Benchmarked 6 model families (linear, RF, SVR, MLP, GPR, TabPFN) on 15 feature groups: R² 0.91 on random splits, 0.90 with real-time-feasible features only
• Showed with leave-one-participant-out and transfer experiments that unseen users drop to R² ≈ 0.5, and found user clusters that motivate the thesis
• Detected and removed target leakage from a physics-derived feature (τ) that had produced a misleading R² = 1.00

SUSTech, Prof. Chenglong Fu · Shenzhen, China

• Built IMU-based gait perception: zero-velocity-update foot tracking, a combined LIPM + ZUPT foothold predictor, and real-time phase-angle gait estimation
• Designed LIPM-based foot trajectories, inverse kinematics and phase-synchronised position control; verified in Simulink
• Achieved stable 10-minute walking trials on the ≈ 15 kg prototype in synchronous and diagonal gaits

TU Delft · Delft, Netherlands · Team project

• Built a ROS/C++ planner from scratch: occupancy mapping, k-PRM with KD-tree, A*, corridor-based trajectory optimisation (OSQP)
• Benchmarked against RRT variants in forest-like and low-connectivity environments

TU Delft · Delft, Netherlands · Team project

• Designed a linear MPC on linearised quadrotor dynamics with a DARE-based terminal cost for stability and feasibility
• Validated across prediction horizons and constraint sets in 3D simulation (Python)

Steam Workshop · Online · Solo

• Designed weighted decision rules over terrain, culture, government and dynasty, tuned to balance plausibility against randomness
• 14,400+ subscribers and 178 overwhelmingly positive ratings; #14 most popular CK3 mod of the past year (Sep 2026), five months after release

Team Epoch × TNO · Netherlands · Solo

• Ranked 26th of 89 teams (215 participants) on the organisers' final hidden test set with an ensemble classifier, working solo
• Designed a radar-first system: calibrated species classifier → collision-risk score → rule-based, auditable curtailment modes