How to improve planning speed?

[allan2013]

My GPU is RTX 5070Ti.
I use plan_single() of goal_pose for single UR10e. I used collsion_mesh_scene.yml provided in the world folder as my world file.
The configuration of parameter are set like the below.

    self.motion_gen_config = MotionGenConfig.load_from_robot_config(
        "ur10e.yml",
        "collsion_mesh_scene.yml",
        self.tensor_args,
        interpolation_dt=0.02,
        use_cuda_graph=True,
        collision_checker_type=MESH,
        num_trajopt_seeds=6,
        num_ik_seeds=30,
        evaluate_interpolated_trajectory=True,
        trajopt_dt=0.5,
        grad_trajopt_iters=500
    )

The planning time I got is around 0.33 seconds which is much slower than the introduction in the homepage. I have never checked the plan_single_js() version which might be faster because of no computation for IK. Do anybody have idea for how to speedup in my case?

[balakumar-s]

grad_trajopt_iters=500 is 350 more iterations than the default value. Can you comment out this parameter and see what planning time you get?

One way to verify the planning time we report in the website is to run the motion planning benchmark provided in: https://curobo.org/get_started/4_benchmarks.html#running-benchmarks