Unconstrained IK (TRAC-IK)¶
TRAC-IK wraps the KDL-derived TRAC-IK C++ solver. It runs a sequential-quadratic-programming solver and a Newton pseudoinverse solver concurrently and returns whichever finds a valid solution first, with optional weighting of the solution by distance, manipulability, or raw speed.
Minimal example¶
import numpy as np
from wheelchair_planning.wheelchair import HOME_JOINTS, JOINT_GROUPS
from wheelchair_planning.kinematics import create_ik_solver
from wheelchair_planning.types import IKConfig, SE3Pose, SolveType
solver = create_ik_solver(
"arm",
config=IKConfig(
timeout=0.2, # seconds per attempt
epsilon=1e-5, # convergence tolerance
solve_type=SolveType.SPEED,
max_attempts=10,
position_tolerance=1e-4,
orientation_tolerance=1e-4,
),
)
home = HOME_JOINTS[JOINT_GROUPS["arm"]]
ee = solver.fk(home) # forward kinematics
target = SE3Pose(
position=ee.position + np.array([0.05, 0.0, -0.05]),
rotation=ee.rotation,
)
result = solver.solve(target, seed=home)
print(result.status, result.joint_positions)
Solve types¶
IKConfig.solve_type picks the post-solution objective when multiple
valid solutions exist in the timeout window:
SolveType |
Picks |
|---|---|
SPEED |
whichever solver finishes first |
DISTANCE |
solution closest to the seed |
MANIP1 |
solution with highest manipulability (Jacobian SVD product) |
MANIP2 |
solution with highest isotropy (smallest / largest Jacobian singular value) |
When to use¶
- Pick-and-place, reach-to-grasp, visual servoing — one end-effector pose, no extra task-space invariants.
- Fast enough for online use — default 0.2 s timeout is conservative; most calls return in a few milliseconds.
- Seeded batches — pass a seed close to the current joint state to get smooth solution continuity across frames.