---

name: ik description: "Solves single, batched, collision-aware, and differentiable inverse-kinematics goals with cuRobo's CUDA IKSolver API." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Inverse kinematics

Use this route for pose IK, reachability sweeps, multi-seed solving, collision- free goals, or updating the world without rebuilding a solver. Read api-reference.md for signatures and workflows.md for recipes.

Core workflow

1. Validate CUDA and choose a robot/tool frame. Build an InverseKinematicsCfg with create(robot=..., num_seeds=...); add scene_model, self_collision_check, collision_cache, and max_batch_size when needed.
2. Construct Pose tensors with (B,3) positions and (B,4) wxyz quaternions, then wrap them as GoalToolPose.from_poses({tool: pose}).
3. Call ik.solve_pose(goal, current_state=..., seed_config=...). Inspect result.success, result.js_solution, position/orientation error, and convergence before accepting a state.
4. For batches, allocate exactly the configured maximum shape and use the success mask; one unreachable pose must not invalidate the whole batch.
5. Call ik.update_world(Scene(...)) for runtime obstacle changes. Use the bounded scripts/ik_smoke.py helper for a headless check; reserve Viser, differential, and reachability viewers for a deliberate UI process.

Read troubleshooting.md for target reachability, shape, quaternion, collision-cache, and CUDA graph failures. Continue to motion-planning when IK is only the first stage of a collision-free path.