Closed cremebrule closed 1 month ago
If it’s not a bug, please use discussions: https://github.com/NVlabs/curobo/discussions
Please provide the below information in addition to your issue:
Running a batch trajectory solve results in the following error, using the latest curobo installed from source:
File ~/projects/repo/curobo/src/curobo/curobolib/geom.py:182, in get_pose_distance(out_distance, out_position_distance, out_rotation_distance, out_p_vec, out_q_vec, out_idx, current_position, goal_position, current_quat, goal_quat, vec_weight, weight, vec_convergence, run_weight, run_vec_weight, offset_waypoint, offset_tstep_fraction, batch_pose_idx, batch_size, horizon, mode, num_goals, write_grad, write_distance, use_metric, project_distance) 169 if batch_pose_idx.shape[0] != batch_size: 170 raise ValueError("Index buffer size is different from batch size") 172 r = geom_cu.pose_distance( 173 out_distance, 174 out_position_distance, 175 out_rotation_distance, 176 out_p_vec, 177 out_q_vec, 178 out_idx, 179 current_position, 180 goal_position.view(-1), 181 current_quat, --> 182 goal_quat.view(-1), 183 vec_weight, 184 weight, 185 vec_convergence, 186 run_weight, 187 run_vec_weight, 188 offset_waypoint, 189 offset_tstep_fraction, 190 batch_pose_idx, 191 batch_size, 192 horizon, 193 mode, 194 num_goals, 195 write_grad, 196 write_distance, 197 use_metric, 198 project_distance, 199 ) 201 out_distance = r[0] 202 out_position_distance = r[1] RuntimeError: view size is not compatible with input tensor's size and stride (at least one dimension spans across two contiguous subspaces). Use .reshape(...) instead.
The fix seems trivial -- simply changing the offending line to be .reshape() seems to fix things.
.reshape()
reshape causes a memory copy which we want to avoid to be more efficient. For the most common case, view would work.
I would move the reshape to be before calling the solve function.
If it’s not a bug, please use discussions: https://github.com/NVlabs/curobo/discussions
Please provide the below information in addition to your issue:
Issue Details
Running a batch trajectory solve results in the following error, using the latest curobo installed from source:
The fix seems trivial -- simply changing the offending line to be
.reshape()
seems to fix things.