Vedant Anand

Abstract

This research investigates whether a constrained model predictive controller (MPC) can improve vehicle yaw stability during a sine-with-dwell manoeuvre when compared with nocontrol, PD and LQR yaw-moment control strategies. The controller is implemented as a high-level direct-yaw-moment controller, while the closed-loop response is evaluated on a nonlinear 9-DoF vehicle model. A simplified three-state bicycle model is used only inside the MPC prediction problem. The study considers manoeuvres at 60kmh−1 and 100kmh−1, evaluating yaw-rate tracking, sideslip and actuator saturation. A prediction-model mismatch study is also performed by varying the predicted vehicle mass and tire cornering stiffness. The results show that the MPC prevents the severe instability observed without control and improves yaw-rate RMS error, sideslip and yaw-moment saturation at 100kmh−1. At 60kmh−1, the MPC and LQR achieve comparable yaw-rate RMS performance. The mismatch study shows only minor variations in yaw-rate tracking and sideslip, with no solver-feasibility issues, indicating that the selected MPC formulation is robust to moderate prediction model errors.

TU Delft course project | Vehicle Dynamics and Control

Abstract

Teledermatology mainly relies on visual information, which can limit the assessment of tissue properties that are not directly visible from images alone. This exploratory study aims to design a device to investigate whether low-cost force--depth measurements can provide an additional mechanical feature for distinguishing scar tissue from healthy skin. Force--depth measurements were performed on scar tissue and healthy skin samples. For each measurement, stiffness was estimated from a linear fit to selected valid data points in the unloading phase. The resulting stiffness values for scar tissue were compared with corresponding healthy-skin reference values. A G* Power analysis was performed to estimate the required sample size, resulting in a required sample size of 22, which was obtained. The comparison shows a statistically significant difference between scar tissue and healthy skin after conducting paired t-test based analysis. Further research can focus on increasing sample size, improve control, or focus on additional feature extraction to determine if this approach can reliably distinguish scar tissue and healthy skin. The preliminary findings show a statistically significant difference between healthy skin tissue and scar tissue, indicating the low-cost option is a promising supplement to diagnoses.

TU Delft course project | Applied Experimental Methods

Abstract

Radar is an attractive sensor for autonomous driving due to its robustness to weather, long detection range, and direct velocity measurement via the Doppler effect. However, radar point clouds are sparse - typically 300 to 500 points per frame, which causes pillar-based detectors to miss small objects such as pedestrians. We present MultiZBEV, a radar-camera fusion framework that addresses this sparsity through two contributions. First, we accumulate radar returns across five temporal frames with ego-motion compensation and spatiotemporal clutter filtering, increasing point cloud density while suppressing false detections. Second, we project camera features from a MobileNetV3-Small backbone into bird's-eye-view space across three height planes, fusing them with radar BEV features via a zero-initialized residual connection. We evaluate on the View of Delft dataset and report a driving corridor mAP of 68.65%, compared to 47.78% for the radar-only baseline. Pedestrian AP improves by 20.8 points, the largest gain across all classes.

TU Delft course project | Advanced Machine Perception

Abstract

A model predictive control (MPC) approach is implemented for the ego vehicle to re-adjust it's behavior with respect to the lead vehicle, in order to achieve cruise like trajectories. We expand on the regulation based MPC using reference tracking and compare with the LQR controller design. Further improvement in the robustness of the system is achieved using an offset free disturbance model.

TU Delft course project | Model Predictive Control (SC42125)

Abstract

A sampling-based planning and control approach is implemented for a car-like robot operating in cluttered environments. A collision-free reference path is generated using RRT*, incorporating non-holonomic constraints through a kinematic bicycle model. The generated trajectory is then tracked using PID and Model Predictive Control (MPC), and their performance is compared under identical constraints. While PID enables fast and reliable tracking, MPC produces smoother and more dynamically consistent motion.

TU Delft course project | Planning and Decision Making (RO47005)

Abstract

An Artificial Intelligence (AI)-based Autonomous Underwater Vehicle (AUV) is developed to enable autonomous navigation, real-time anomaly detection, and debris differentiation in underwater environments. A systems engineering and mechatronics approach is adopted to design a compact, modular structure with integrated insulation and waterproofing for improved resilience. Sensor fusion techniques are implemented for environmental perception and data acquisition, while machine learning algorithms are used for navigation and anomaly detection. The system is validated through subsystem integration and testing, demonstrating enhanced maneuverability and reliable operation in challenging conditions.

Multidisciplinary Project | Team SRM AUV

Abstract

An articulated robotic arm is designed with improved motion control and reduced back-drivability using cycloidal drives in the shoulder and elbow joints, replacing conventional linear actuators. A multi-DOF end effector based on a triple bevel gear configuration is integrated to achieve yaw, pitch, and roll motion. A belt-driven gear reduction system is proposed to regulate motor speed and enhance torque for precise gripper operation. The design demonstrates improved stability, control, and manipulation capability.

University Rover Challenge (URC) | Team Rudra

Abstract

System Design and Integration of Soil Sampling, Collection (Using Rack and Pinion Based Auger System) and Retrieval. Drive Mechanism formulated with special rubber materials withstanding high temperatures, and adapting multithreaded tyre design along with differential drive for mobility in terrestrial hydrothermal conditions.

Multidisciplinary Project | Team Rudra

Abstract

A compact produce chopping mechanism is designed to enable controlled and adaptable processing of fruits and vegetables. The system incorporates a flow-based actuation concept to regulate the movement of produce, along with a blade selection mechanism to accommodate different user-defined cutting requirements. A linear actuator is integrated to facilitate vertical chopping operations, enhancing the range of achievable cutting patterns. The design focuses on mechanical simplicity, adaptability, and repeatable performance across varying produce types.

Navyantram | National Robotics Competition

Abstract

A snow-clearing vehicle is designed to address the challenges of snow removal in high-altitude areas. The system integrates a scooper and impeller-based mechanism for snow collection and displacement. A tracked belt drive system is selected to improve traction and mobility over snow-covered terrain. An additional attachable sprinkling system is incorporated to enable uniform distribution of eco-friendly chemicals, reducing snow accumulation and improving overall clearing efficiency.

Defense Service Hackathon | SRM University

Abstract

Compact Extraterrestrial Navigation Technology for Autonomous Underground Reconnaissance(CENTAUR) is a hybrid mobility rover designed to enable efficient navigation across uneven and obstacle-rich terrains. The system integrates a dual-mode locomotion mechanism, allowing operation as a four-wheeled drive platform for efficient traversal on relatively smooth surfaces, and reconfiguration into a six-legged system employing a tripod gait for obstacle negotiation and climbing. The mechanical and drive architecture is developed to ensure reliable torque transmission, stability, and adaptability under varying terrain conditions. The overall design demonstrates improved mobility and versatility by combining the speed of wheeled locomotion with the terrain-handling capability of legged systems.

International Rover Design Challenge (IRDC) | Team RUDRA