Pros and cons Human-like maps Exploration strategy built into mapping process Purely on-line Con: Loop closure Will not necessarily map space completely Absence of decision -theoretic measures means no measure of map quality a large number of free parameters and heuristics Purely on-line The Extended Kalman Filter for Building Maps Process model is now x=f(x1-1,l1-1,q=1) Measurement model is h(x1,r;) Linearising, we get x,=f(x1-1,1-1,0) x≈x1+A(x1-1-x1)+Wq1 二≈h(x1,0)+H(x1-x,)+V1 Where A.H. w and v are the jacobians ah ch afr, a af, ahPros and Cons ● Pro:● Human-like maps ● Exploration strategy built into mapping process ● Purely on-line ● Con:● Loop closure ● Will not necessarily map space completely ● Absence of decision-theoretic measures means no measure of map quality ● A large number of free parameters and heuristics ● Purely on-line ● Process model is now ● Measurement model is ● Linearising, we get ● where A, H, W and V are the Jacobians: The Extended Kalman Filter for Building Maps ( , , ) t = t−1 t−1 t−1 x f x u q ( , ) t t t z = h x r t t t t t t t t t t t t t z h x H x x Vr x x A x x Wq x f x u ≈ + − + ≈ + − + = − − − − − ) ~ ,0) ( ~( ( ˆ ) ~ ( , ,0) ~ 1 1 1 1 1 ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ = n n n n x f x f x f x f A L M O M L 1 1 1 1 ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ = n m m n x h x h x h x h H L M O M L 1 1 1 1 ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ = n n n n q f q f q f q f W L M O M L 1 1 1 1 ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ ⎤ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ⎡ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ = m m m m v h v h v h v h V L M O M L 1 1 1 1