Trajectory of motion generation

[Entongsu]

Hi,

I've encountered trajectory generation for the Franka robot. During recent operations, I observed that the robot consistently overshoots the intended target position and then corrects its path by moving backward. I am not sure how to fix this problem. Could you give me some hints over that?

[balakumar-s]

What commit of curobo are you using?

[Entongsu]

I think I have used this version 1e0b5a80c295fb91bc4af97dc219504cba8a6f16. I have tried the newest one and got the error of error[~success] += 10000.0 RuntimeError: CUDA error: an illegal memory access was encountered CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect. For debugging consider passing CUDA_LAUNCH_BLOCKING=1. Compile withTORCH_USE_CUDA_DSAto enable device-side assertions.

[Entongsu]

I have tried version 3bfed9d7737fc3865595e4a50b1fe84a32cf627f. The generated trajectories are worse than before.

[balakumar-s]

Are you using batched planning or single pose planning?

[Entongsu]

I use single-pose planning. Could you provide some recommended settings for the parameter? Thank you. ` motion_gen_config = MotionGenConfig.load_from_robot_config( robot_file, world_file, self.tensor_args,

trajopt_tsteps=32,

        # collision_checker_type=CollisionCheckerType.PRIMITIVE,
        # use_cuda_graph=True,
        # num_trajopt_seeds=4,
        # num_graph_seeds=4,
        # evaluate_interpolated_trajectory=True,
        interpolation_dt=0.01,
    )`

[Entongsu]

I have found the problem; it may stem from the collision mesh I used for the motion generation. I have set something like this before. cuboids = [ Cuboid(name="cube", pose=np.concatenate([ translation_r2cube, quaternion_r2cube[[3, 0, 1, 2]] ]), dims=np.array(self.env.cube.size)*2) ]. After I change the dims from np.array(self.env.cube.size)*2 to np.array(self.env.cube.size), it becomes much more better. For the dims of cuboids, is it the full size or half size of the cube?

[balakumar-s]

Dims is full size. Also make sure finetune_trajopt is enabled in MotionGenPlanConfig.

[balakumar-s]

Are you still having issues? If yes, please share a small script that we can reproduce. Closing for now.