Jonas Fischer | Academic Projects

LVSM-VAE: Transformer-Based Novel View Synthesis in Latent Space

Trains LVSM in a VAE latent space to reduce inference cost while retaining novel-view quality and enabling longer context sequences under fixed compute budgets.

Novel View Synthesis Transformers VAE Computer Vision

Self-Supervised Learning for Tactile Classification

Explores self-supervised learning for tactile sensing with a shared TacNet-II encoder and time-based masking, improving accuracy for material classification.

Self-Supervised Tactile Sensing Material Classification

A Multi-objective Approach to Safe Reinforcement Learning for Smart Grids

Frames smart grid control as multi-objective RL with separate cost and safety objectives, learning Pareto-optimal trade-offs.

Reinforcement Learning Smart Grids Optimization

Parameterized and Dynamic Gaussian Mixture Models for Robot Manipulation

Compares task-parameterized GMMs and elastic dynamical systems for adapting robot motions from demonstrations to new task parameters.

Robot Manipulation Gaussian Mixture Models Trajectory Learning