From e3dcf60b3ea0c5d038331df67498049c919527df Mon Sep 17 00:00:00 2001 From: Rick Staa <rick.staa@outlook.com> Date: Mon, 4 Oct 2021 13:59:24 +0200 Subject: [PATCH] Add TrajOpt planner config This commit adds the files needed for running the trajopt planner example that is found in the [moveit_tutorials](https://ros-planning.github.io/moveit_tutorials/doc/trajopt_planner/trajopt_planner_tutorial.html). This planner is not included into `move_group.launch` by default but can be invoked via the `pipeline` argument. This was done since it is not yet officially released (see https://github.com/ros-planning/moveit/issues/1467). --- config/trajopt_planning.yaml | 58 +++++++++++++++++++++ launch/run_benchmark_trajopt.launch | 21 ++++++++ launch/trajopt_planning_pipeline.launch.xml | 32 ++++++++++++ 3 files changed, 111 insertions(+) create mode 100644 config/trajopt_planning.yaml create mode 100644 launch/run_benchmark_trajopt.launch create mode 100644 launch/trajopt_planning_pipeline.launch.xml diff --git a/config/trajopt_planning.yaml b/config/trajopt_planning.yaml new file mode 100644 index 0000000..6c6ea43 --- /dev/null +++ b/config/trajopt_planning.yaml @@ -0,0 +1,58 @@ +trajopt_param: + improve_ratio_threshold: 0.25 # minimum ratio true_improve/approx_improve to accept step. [Research Paper] TrueImprove/ModelImprove > c + min_trust_box_size: 1e-4 # if trust region gets any smaller, exit and report convergence. [Research Paper] xtol + min_approx_improve: 1e-4 # if model improves less than this, exit and report convergence + min_approx_improve_frac: -.Inf # if model improves less than this, exit and report convergence + max_iter: 100 # The max number of iterations + trust_shrink_ratio: 0.1 # if improvement is less than improve_ratio_threshold, shrink trust region by this ratio. [Research Paper] tao- + trust_expand_ratio: 1.5 # if improvement is less than improve_ratio_threshold, shrink trust region by this ratio. [Research Paper] tao+ + cnt_tolerance: 1e-4 # after convergence of penalty subproblem, if constraint violation is less than this, we're done. [Research Paper] ctol + max_merit_coeff_increases: 5 # number of times that we jack up penalty coefficient. [Reasearch Paper] Max iteration in PenaltyIteration loop + merit_coeff_increase_ratio: 10 # ratio that we increate coeff each time. [Researcy Paper] Penalty scaling factory (italic kappa) + max_time: .inf # not yet implemented + merit_error_coeff: 10 # initial penalty coefficient. [Research Paper] mu_0 + trust_box_size: 1e-1 # current size of trust region (component-wise). [Research Paper] s + +problem_info: + basic_info: + n_steps: 20 # 2 * steps_per_phase + dt_upper_lim: 2.0 # The upper limit of 1/dt values allowed in the optimization + dt_lower_lim: 100.0 # The lower limit of 1/dt values allowed in the optimization + start_fixed: true # if true, start pose is the current pose at timestep = 0 + # if false, the start pose is the one given by req.start_state + use_time: false # if false, it means the timestep is unit, meaning x1-x0 is the velocity for example + # if true, then cost_infos ro cnt_infos should have TT_USE_TIME for their term_type + convex_solver: 1 # which convex solver to use + # 1 = AUTO_SOLVER, 2 = BPMPD, 3 = OSQP, 4 = QPOASES, 5 = GUROBI + + init_info: + type: 3 # 1 = STATIONARY, 2 = JOINT_INTERPOLATED, 3 = GIVEN_TRAJ + # request.start_state should be provided for JOINT_INTERPOLATED and GIVEN_TRAJ + dt: 0.5 + +joint_pos_term_info: + start_pos: # from req.start_state + name: start_pos + first_timestep: 10 # First time step to which the term is applied. + last_timestep: 10 # Last time step to which the term is applied. + # if start_fixed is trure then we can not set the first_timestep to 0 for start_pos as it is going + # to be ignored and only the current pose would be the constraint at timestep = 0. + term_type: 2 # 1 = TT_COST, 2 = TT_CNT, 3 = TT_USE_TIME + middle_pos: + name: middle_pos + first_timestep: 15 + last_timestep: 15 + term_type: 2 + goal_pos: + name: goal_pos + first_timestep: 19 + last_timestep: 19 + term_type: 2 + goal_tmp: # If the constraint from request does not have any name, for example when using MotionPlanning Display in Rviz, + # goal_tmp is assigned as the name of the constraint. + # In this case: start_fixed = false and start_pos should be applied at timestep 0 + # OR: start_fixed = true and start_pos can be applies at any timestep + name: goal_tmp + first_timestep: 19 # n_steps - 1 + last_timestep: 19 # n_steps - 1 + term_type: 2 diff --git a/launch/run_benchmark_trajopt.launch b/launch/run_benchmark_trajopt.launch new file mode 100644 index 0000000..7628416 --- /dev/null +++ b/launch/run_benchmark_trajopt.launch @@ -0,0 +1,21 @@ +<launch> + + <!-- This argument must specify the list of .cfg files to process for benchmarking --> + <arg name="cfg" /> + + <!-- Load URDF --> + <include file="$(find moveit_resources_panda_moveit_config)/launch/planning_context.launch"> + <arg name="load_robot_description" value="true"/> + </include> + + <!-- Start the database --> + <include file="$(find moveit_resources_panda_moveit_config)/launch/warehouse.launch"> + <arg name="moveit_warehouse_database_path" value="moveit_trajopt_benchmark_warehouse"/> + </include> + + <!-- Start Benchmark Executable --> + <node name="$(anon moveit_benchmark)" pkg="moveit_ros_benchmarks" type="moveit_run_benchmark" args="$(arg cfg) --benchmark-planners" respawn="false" output="screen"> + <rosparam command="load" file="$(find moveit_resources_panda_moveit_config)/config/trajopt_planning.yaml"/> + </node> + +</launch> diff --git a/launch/trajopt_planning_pipeline.launch.xml b/launch/trajopt_planning_pipeline.launch.xml new file mode 100644 index 0000000..ec9ea9b --- /dev/null +++ b/launch/trajopt_planning_pipeline.launch.xml @@ -0,0 +1,32 @@ +<launch> + + <!-- TrajOpt Plugin for MoveIt --> + <arg name="planning_plugin" value="trajopt_interface/TrajOptPlanner" /> + + <!-- define capabilites that are loaded on start (space seperated) --> + <arg name="capabilities" default=""/> + + <!-- inhibit capabilites (space seperated) --> + <arg name="disable_capabilities" default=""/> + + <!-- The request adapters (plugins) used when planning with TrajOpt. + ORDER MATTERS --> + <arg name="planning_adapters" default="default_planner_request_adapters/AddTimeParameterization + default_planner_request_adapters/FixWorkspaceBounds + default_planner_request_adapters/FixStartStateBounds + default_planner_request_adapters/FixStartStateCollision + default_planner_request_adapters/FixStartStatePathConstraints" /> + + <arg name="start_state_max_bounds_error" default="0.1" /> + <arg name="jiggle_fraction" default="0.05" /> + + <param name="planning_plugin" value="$(arg planning_plugin)" /> + <param name="request_adapters" value="$(arg planning_adapters)" /> + <param name="start_state_max_bounds_error" value="$(arg start_state_max_bounds_error)" /> + <param name="jiggle_fraction" value="$(arg jiggle_fraction)" /> + <param name="capabilities" value="$(arg capabilities)" /> + <param name="disable_capabilities" value="$(arg disable_capabilities)" /> + + <rosparam command="load" file="$(find panda_moveit_config)/config/trajopt_planning.yaml"/> + +</launch> -- GitLab