60 lines
2.3 KiB
Python
60 lines
2.3 KiB
Python
#
|
|
# Copyright (c) 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
#
|
|
# NVIDIA CORPORATION, its affiliates and licensors retain all intellectual
|
|
# property and proprietary rights in and to this material, related
|
|
# documentation and any modifications thereto. Any use, reproduction,
|
|
# disclosure or distribution of this material and related documentation
|
|
# without an express license agreement from NVIDIA CORPORATION or
|
|
# its affiliates is strictly prohibited.
|
|
#
|
|
|
|
# Third Party
|
|
import pytest
|
|
import torch
|
|
|
|
# CuRobo
|
|
from curobo.types.base import TensorDeviceType
|
|
from curobo.types.robot import JointState
|
|
from curobo.wrap.reacher.motion_gen import MotionGen, MotionGenConfig, MotionGenPlanConfig
|
|
from curobo.wrap.reacher.evaluator import TrajEvaluatorConfig
|
|
|
|
@pytest.fixture(scope="module", params=[True, False])
|
|
def evaluate_interpolated_trajectory(request):
|
|
return request.param
|
|
|
|
@pytest.fixture(scope="module")
|
|
def motion_gen(evaluate_interpolated_trajectory):
|
|
tensor_args = TensorDeviceType()
|
|
world_file = "collision_test.yml"
|
|
robot_file = "franka.yml"
|
|
dof = 9
|
|
traj_evaluator_config = TrajEvaluatorConfig(
|
|
max_acc=torch.ones((dof), device=tensor_args.device, dtype=tensor_args.dtype),
|
|
max_jerk=torch.ones((dof), device=tensor_args.device, dtype=tensor_args.dtype),
|
|
min_dt=torch.tensor(0.01, device=tensor_args.device, dtype=tensor_args.dtype),
|
|
max_dt=torch.tensor(1.5, device=tensor_args.device, dtype=tensor_args.dtype)
|
|
)
|
|
motion_gen_config = MotionGenConfig.load_from_robot_config(
|
|
robot_file,
|
|
world_file,
|
|
tensor_args,
|
|
trajopt_tsteps=26,
|
|
use_cuda_graph=False,
|
|
num_trajopt_seeds=50,
|
|
fixed_iters_trajopt=True,
|
|
evaluate_interpolated_trajectory=evaluate_interpolated_trajectory,
|
|
traj_evaluator_config=traj_evaluator_config,
|
|
)
|
|
motion_gen = MotionGen(motion_gen_config)
|
|
motion_gen.warmup(warmup_js_trajopt=True)
|
|
|
|
retract_cfg = motion_gen.get_retract_config()
|
|
|
|
start_state = JointState.from_position(retract_cfg.view(1, -1).clone())
|
|
goal_state = JointState.from_position(retract_cfg.view(1, -1).clone())
|
|
|
|
result = motion_gen.plan_single_js(start_state, goal_state, MotionGenPlanConfig(max_attempts=1))
|
|
|
|
def test_motion_gen(motion_gen):
|
|
return True |