U=diag([0.05^2 (5*pi/180)^2]);

zu=2.68; zv=3.06; zd=1.30;

Missing observation causes

Warning: Matrix is close to singular or badly scaled.
         Results may be inaccurate. RCOND = 1.062233e-37.

Kinematic constraints are classically

divided into two classes: constraints

that can be integrated to yield constraints

on the position variables,

called holonomic constraints, and constraints

for which this integration is

not possible, called nonholonomic constraints.

A typical example of a nonholonomic

constraint is a wheel rolling

vertically without slipping on a surface.

The constraint on the allowable velocity

(the point of contact of the wheel

with the surface cannot slip in all

directions) cannot be integrated to

yield a constraint on the position of

the wheel. This nonintegrability is

intuitively clear, as illustrated by the

fact that an automobile can go anywhere

it pleases by suitable maneuvering.

(This example hints at the

intimate relationship between nonholonomic

constraints and controllability.)

Loosely speaking, mechanical

systems with holonomic constraints

can be reduced to lower dimensional

mechanical systems without constraints;

for systems with nonholonomic

constraints, this reduction is not

possible, and as a result some distinguishing

features arise

both cubicle simulations and the small circle shows bias if the sensors were assumed slightly out of alignment.

Noticed when running simulation results on fullslam3() and rippleSLAM() the B/2 correction usually used caused large bias resulting in inconsistency.

Using Treemap as a Generic Least Square - Backend for 6-DOF SLAM:

Udo Frese -> talks of large maps in 3D in less than a sec using form of LS.

• fullslam to use uvd and preprocessed data.
• Sanitizing a data set.
•  Error=Estimate-Truth
• What's an exact solution?
• Non holonomic constraint