ABSTRACT
This chapter deals with the problem of evaluating how much the rotation computed from sensor data is accurate. For this, the authors consider both experimental and theoretical evaluation. Experimental evaluation deals with the situation where a large number of values computed in similar circumstances or simulation results are obtained. Finally, they point out that there exists a theoretical accuracy bound on the covariance matrix, called the “Kanatani–Cramer–Rao lower boud,” whatever computational algorithms are used. They show that maximum likelihood computation achieves that bound up to high order noise terms. This observation is not limited to maximum likelihood estimation; this generally applies when parameter estimation is reduced to minimization of some function.
