Multi-modal perception for soft robotic interactions using generative models