We propose to develop data-driven outlier-robust state estimation of model-free processes. This would be an extension of our developed paper DNS (presented at ICASSP 2026) and out two manuscripts DSR and RoVaS (both submitted to NeurIPS). Our first objective is the development of outlier-robust theoretical framework for sequential monte carlo (SMC). This will be studied in the variational SMC (VSMC). Further, we will study multi-object tracking, also for model-free processes. These two problems are highly connected, as both require modelling of unreliable or noisy measurements. My supervisor is Saikat Chatterjee, professor at Information Science and Engineering at KTH.