A multi-modal generative model to forecast the future positions of agents in traffics.

Programming contest hosted by CV Zone, sponsored by NVIDIA. Where the objective is to predict the paths of the ball on the pool table and whether they will go in the hole.

I made an educational Rocket Trajectory Simulation Assignment

Implement deep learning based trajectory prediction methods

Polar Collision Grids: Effective Interaction Modelling for Pedestrian Trajectory Prediction in Shared Space Using Collision Checks (ITSC 2023)

We introduce a novel trajectory predictor that considers social interactions among agents, maintaining spatial-temporal information over an extended temporal horizon. It achieves high accuracy, generalisability, and outperforms recent SOTA algorithms on NGSIM and HIGHD Datasets

Code for Bachelor's thesis: Observation of humans using computer vision for autonomous driving

The project involves building a scenario consisting of a Thrower and a Goalie robot arms. The thrower moves to the desired position and throws the ball while the task of the goalie is to intercept and block the incoming ball from making a goal.

Official code for AAAI 2023 paper "Multi-stream Representation Learning for Pedestrian Trajectory Prediction"

Code for "Transformer Networks for 4D Trajectory Prediction"

Modularized, annotated and extended code for LaPred: Lane-Aware Prediction of Multi-Modal Future Trajectories of Dynamic Agents by Kim et al.’s, based on their original implementation.

FIR-based trajectory prediction at nighttime

[T-ITS 2024] TrajPRed: Trajectory Prediction with Region-based Relation Learning

Prädiktion von zukünftigen Fußgänger-Trajektorien/Bewegungsbahnen mithilfe eines LSTM-Neuronalen Netzes.

Vehicle-Pedestrian Motion Prediction using Classification

[under review] Official PyTorch Implementation of "Multi-scale Feature Aggregation Autoregressive Network for Multi-Agent Trajectory Prediction".

Code for "CARPe Posterum: A Convolutional Approach for Real-time Pedestrian Path Prediction"

We have compared 4 models- Vanilla LSTM, Social LSTM, OLSTM, and GRU to show their comparison for predicting non linear trajectories of pedestrians in different scenes. We demonstrate their performance on publically available datasets. We show how it is important to take into account the surroundings of the pedestrians to have a better accuracy.