With the advent of more and more dynamic motions in less structured environments, the forces exerted on the robots become not only less predictable but also less modelable. In this context, the robots are required to provide at least three guarantees. The first one is to limit these forces for the sake of safety, and thus to estimate their positions and intensities. We show in this talk how pushing beyond the quest for the mere precision of the floating-based kinematics we can build a rich estimation of the kinetics and the external forces applied on the robot. The second requirement is to prevent failure and to be robust to these disturbances. We present for that an estimator theoretically guaranteeing to fully ignore such disturbances whatever their magnitude. The third requirement is to guarantee a sufficient level of performance despite the disturbance. To realize that we show how to achieve passive compliance with minimal task-performance loss even on robots with a high reduction ratio and no torque feedback. Finally, we discuss the important missing bricks to actually respect these requirements.