Continual learning-based trajectory prediction with memory augmented networks