RRT* in configuration spaces (single integrator) with trajectory biasing heuristic. This file provides the code for running the RRT* algorithm for a single integrator with bounds on velocity, i.e., planning in configuration spaces using a trajectory biasing heuristic, which concentrates the samples around the current best trajectory in the RRT*. The system can be run in arbitrary dimensions.
The RRT* algorithm scales well beyond 20 dimensions (upto 50) converging close to an optimal solution in a few seconds, in the examples provided in this file when using the trajectory biasing heuristic.
The dimensionality of the space can be adjusted using the NUM_DIMENSIONS parameter. Also the trajectory biasing parameters can be varied from the within the main function.
// Standard header files #include<iostream> using namespace std; // SMP HEADER FILES ------ #include <smp/components/samplers/trajectory_bias.hpp> #include <smp/components/distance_evaluators/kdtree.hpp> #include <smp/components/extenders/single_integrator.hpp> #include <smp/components/collision_checkers/standard.hpp> #include <smp/components/multipurpose/minimum_time_reachability.hpp> #include <smp/planners/rrtstar.hpp> #include <smp/planner_utils/trajectory.hpp> // PARAMETERS TO THE PROBLEM *********************************************************************************** // * #define NUM_DIMENSIONS 10 // Change the number of dimensions from here. Scale it up to 20 - 30 // dimensions to see the convergence of RRT* towards an optimal solution // in very high dimensional configuration spaces without employing any heuristics. #define EXTENSION_LENGTH 25.0 // Maximum length of an extension. This parameter should ideally // be equal longest straight line from the initial state to // anywhere in the state space. In other words, this parameter // should be "sqrt(d) L", where d is the dimensionality of space // and L is the side length of a box containing the obstacle free space. // __NOTE__: Smaller values of this parameter will lead to a good feasible // solution very quickly, whilenot affecting the asymptotic optimality // property of the RRT* algorithm. // * // ************************************************************************************************************* // SMP TYPE DEFINITIONS ------- using namespace smp; // State, input, vertex_data, and edge_data definitions typedef state_single_integrator<NUM_DIMENSIONS> state_t; typedef input_single_integrator input_t; typedef minimum_time_reachability_vertex_data vertex_data_t; typedef minimum_time_reachability_edge_data edge_data_t; // Create the typeparams structure typedef struct _typeparams { typedef state_t state; typedef input_t input; typedef vertex_data_t vertex_data; typedef edge_data_t edge_data; } typeparams; // Define the trajectory type typedef trajectory<typeparams> trajectory_t; // Define all planner component types typedef sampler_trajectory_bias<typeparams,NUM_DIMENSIONS> sampler_t; typedef distance_evaluator_kdtree<typeparams,NUM_DIMENSIONS> distance_evaluator_t; typedef extender_single_integrator<typeparams,NUM_DIMENSIONS> extender_t; typedef collision_checker_standard<typeparams,NUM_DIMENSIONS> collision_checker_t; typedef minimum_time_reachability<typeparams,NUM_DIMENSIONS> min_time_reachability_t; // Define all algorithm types typedef rrtstar<typeparams> rrtstar_t; void *pointer_to_sampler; int wrapper_to_sampler_update_trajectory (trajectory_t *trajectory_in) { ((sampler_t*)(pointer_to_sampler))->update_trajectory (trajectory_in); return 1; } int main () { // 1. CREATE PLANNING OBJECTS // 1.a Create the components sampler_t sampler; distance_evaluator_t distance_evaluator; extender_t extender; collision_checker_t collision_checker; min_time_reachability_t min_time_reachability; // 1.b Create the planner algorithm -- Note that the min_time_reachability variable acts both // as a model checker and a cost evaluator. rrtstar_t planner (sampler, distance_evaluator, extender, collision_checker, min_time_reachability, min_time_reachability); planner.parameters.set_phase (2); // The phase parameter can be used to run the algorithm as an RRT, // See the documentation of the RRG algorithm for more information. planner.parameters.set_gamma (35.0); // Set this parameter should be set at least to the side length of // the (bounded) state space. E.g., if the state space is a box // with side length L, then this parameter should be set to at // least L for rapid and efficient convergence in trajectory space. planner.parameters.set_dimension (NUM_DIMENSIONS); planner.parameters.set_max_radius (EXTENSION_LENGTH); // 2. INITALIZE PLANNING OBJECTS // 2.a Initialize the sampler region<NUM_DIMENSIONS> sampler_support; for (int i = 0; i < NUM_DIMENSIONS; i++) { sampler_support.center[i] = 0.0; sampler_support.size[i] = 20.0; } sampler.set_support (sampler_support); pointer_to_sampler = &sampler; // 2.b Initialize the distance evaluator // Nothing to initialize. One could change the kdtree weights. // 2.c Initialize the extender extender.set_max_length (EXTENSION_LENGTH); // 2.d Initialize the collision checker region<NUM_DIMENSIONS> obstacle_new; for (int i = 0; i < 2; i++) { obstacle_new.center[i] = 5.0; obstacle_new.size[i] = 5.0; } obstacle_new.center[1] = 4.0; for (int i = 2; i < NUM_DIMENSIONS; i++) { obstacle_new.center[i] = 0.0; obstacle_new.size[i] = 20.0; } collision_checker.add_obstacle (obstacle_new); for (int i = 0; i < 2; i++) { obstacle_new.center[i] = -5.0; obstacle_new.size[i] = 5.0; } collision_checker.add_obstacle (obstacle_new); obstacle_new.center[0] = 5.0; obstacle_new.center[1] = -5.0; for (int i = 0; i < 2; i++) { obstacle_new.size[i] = 5.0; } if (NUM_DIMENSIONS >=3 ) { obstacle_new.center[2] = 7.0; obstacle_new.size[2] = 5.0; } collision_checker.add_obstacle (obstacle_new); if (NUM_DIMENSIONS >=3 ) { obstacle_new.center[2] = -7.0; obstacle_new.size[2] = 5.0; } collision_checker.add_obstacle (obstacle_new); obstacle_new.center[0] = -5.0; obstacle_new.center[1] = 5.0; obstacle_new.size[0] = 10.0; obstacle_new.size[1] = 5.0; if (NUM_DIMENSIONS >=3 ) { obstacle_new.center[2] = 0.0; obstacle_new.size[2] = 10.0; } collision_checker.add_obstacle (obstacle_new); // 2.e Initialize the model checker (with minimum_time_reachability). region<NUM_DIMENSIONS> region_goal; for (int i = 0; i < 2; i++) { region_goal.center[i] = 8.0; region_goal.size[i] = 2.0; } for (int i = 2; i < NUM_DIMENSIONS; i++) { region_goal.center[i] = 0.0; region_goal.size[i] = 20.0; } min_time_reachability.set_goal_region (region_goal); // 2.e' Initialize the cost evaluator (with minimum_time_reachability). // --- Register a callback to the sampler class. The callback function is called // whenevert the planner finds a better trajectory that reaches the goal region. min_time_reachability.register_new_update_function (&wrapper_to_sampler_update_trajectory); // 2.f Initialize the planner state_t *state_initial = new state_t; for (int i = 0; i < NUM_DIMENSIONS; i++) { state_initial->state_vars[i] = 0.0; } planner.initialize (state_initial); // 3. RUN THE PLANNER for (int i = 0; i < 5000; i++){ planner.iteration (); if (i%100 == 0){ cout << "Iteration : " << i << endl; } } // 4. GET THE RESULTS trajectory_t trajectory_final; min_time_reachability.get_solution (trajectory_final); return 1; }
