closed
closed constraint motion planning
YeeKal
โข
โข
"#closed"
A Probabilistic Roadmap Approach for Systems with Closed
-
1999
-
break the chain loops
- randomized gradient descent to generate samples
- tangent space sample to connect neighbor points
A Kinematics-Based Probabilistic Roadmap Method for Closed Kinematic Chains
- 2000
-
KBPRM: kinematics based prm
-
node generation
- randomly generate $\theta_a$
- forward kinematics to compute end-frame configuration $g_{la}$
- inverse kinematics to solve $\theta_p$
- $\theta=(\theta_a,\theta_p)$,is retained if it is collision-free
- node connection
Randomized path planning for linkages with closed kinematic chains
- 2001
- randomized gradient descent
random samples
- kinematic error
- random q in $C_{free}$
- compare and select q with smaller error(gradient descend)
Other approaches, such as the Levenberg-Marquardt nonlinear optimization algorithm could be used instead of randomized descent.
GENERATE_RANDOM_SAMPLE()
- q$\gets$ RANDOM_CONFIGURATION();
- $i\gets 0;j\gets 0$
- while $i<I$ and $j
\epsilon$ do - i++; j++;
- $q'\gets \text{RANDOM_NHBR}(q)$;
- if $e(q')<e(q)$ then
- $j\gets 0;q\gets q';$
- if $e(q)\leq \epsilon$ then
- Return $q$
- else
- Return FAILURE
connect points: incremental motions
step is small enough
-
tangent space sample: SVD on the matrix of the partial derivatives to find the orthonormal basis.recursive derivative: then the position could be computed with the derivatives directly:
-
random sample
CONNECT_CONFIGURATIONS(q,q')
- $i\gets 0;j\gets 0; k\gets 0; L\gets {q};$
- while $i<I$ and $j<J$ and$k
\rho_0$ do - i++; j++;
- $q{''}\gets \text{RANDOM_NHBR}(LAST(L))$;
- if $e(q{''})<\epsilon$ then
- $j\gets 0;k++;$
- if $\rho(q{''},q')<\rho(LAST(L),q')$ then
- $k\gets 0; L\gets L+{q{''}};$
- if($\rho (LAST(L),q')\leq \rho_0$) then
- Return L;
- else
- Return FAILURE
tangent space