File constants.hpp

Generates an action based on state and spline (both are cuda globals), returns it. Tunable parameters in and.

namespace controls

Enums

enum class TorqueMode

Values:

enumerator AWD

enumerator FWD

enumerator RWD

Functions

const rclcpp::QoS state_qos (rclcpp::KeepLast(1))

const rclcpp::QoS world_twist_qos (rclcpp::KeepLast(1))

const rclcpp::QoS world_quat_qos (rclcpp::KeepLast(1))

const rclcpp::QoS world_pose_qos (rclcpp::KeepLast(1))

Variables

constexpr const char *controller_node_name = "controller"

constexpr const char *control_action_topic_name = "control_action"

constexpr const char *spline_topic_name = "spline"

constexpr const char *state_topic_name = "state"

constexpr const char *world_twist_topic_name = "filter/twist"

constexpr const char *world_quat_topic_name = "filter/quaternion"

constexpr const char *world_pose_topic_name = "filter/pose"

constexpr const char *controller_info_topic_name = "controller_info"

static const rmw_qos_profile_t best_effort_profile = {RMW_QOS_POLICY_HISTORY_KEEP_LAST, 1, RMW_QOS_POLICY_RELIABILITY_BEST_EFFORT, RMW_QOS_POLICY_DURABILITY_VOLATILE, RMW_QOS_DEADLINE_DEFAULT, RMW_QOS_LIFESPAN_DEFAULT, RMW_QOS_POLICY_LIVELINESS_SYSTEM_DEFAULT, RMW_QOS_LIVELINESS_LEASE_DURATION_DEFAULT, false}

Profile for best effort communication.

const rclcpp::QoS best_effort_qos = rclcpp::QoS(rclcpp::QoSInitialization(best_effort_profile.history, best_effort_profile.depth), best_effort_profile)

const rclcpp::QoS control_action_qos = best_effort_qos

const rclcpp::QoS spline_qos = best_effort_qos

const rclcpp::QoS controller_info_qos = best_effort_qos

constexpr rcl_clock_type_t default_clock_type = RCL_ROS_TIME

constexpr double controller_freq = 10.

Target number of controller steps per second, in Hz. 1 controller step outputs 1 control action.

constexpr float controller_period = 1. / controller_freq

Target duration between control actions, in sec.

constexpr uint32_t num_samples = 1024 * 64

Number of trajectories sampled each controller step.

constexpr uint32_t num_timesteps = 16

Number of controller steps simulated into the future.

constexpr uint8_t action_dims = 2

\(q\), dimensions of Action

constexpr uint8_t state_dims = 4

\(p\), dimensions of State

constexpr float temperature = 1.0f

Convergence speed/stability tradeoff, see LaTeX for more details.

constexpr unsigned long long seed = 0

For brownian pseudo-RNG.

constexpr uint32_t num_action_trajectories = action_dims * num_timesteps * num_samples

Number of elements in the tensor containing all the sampled action trajectories.

constexpr float init_action_trajectory[num_timesteps * action_dims] = {}

Best guess of action trajectory when controller first starts.

constexpr float action_momentum = 0.0f

How much of last action taken to retain in calculation of next action.

constexpr float offset_1m_cost = 5.0f

Cost for being 1m away from midline.

constexpr float target_speed = 10.0f

Linear cost for under target speed, NO cost for above, in m/s.

constexpr float speed_off_1mps_cost = 1.0f

Cost for being 1m/s below target_speed

constexpr float out_of_bounds_cost = 100.0f

Cost for being out of (fake) track bound as defined by track_width.

Reason for not using infinity: reduction uses log of the cost (trading precision for representable range). This covers the edge case where every trajectory goes out of bounds, allowing us to still extract useful information.

constexpr float spline_frame_separation = 0.5f

constexpr uint32_t curv_frame_lookup_tex_width = 512

constexpr float curv_frame_lookup_padding = 0

constexpr float track_width = 30.0f

Not real track width, used for curvilinear frame lookup table generation.

constexpr float car_padding = std::max(spline_frame_separation, M_SQRT2f32 * track_width)

Represents space the car occupies, used to calculate the size of the curvilinear lookup table.

constexpr bool reset_pose_on_spline = true

Sets pose to 0 vector for new spline (sensor POV)

constexpr float cg_to_front = 0.775

constexpr float cg_to_rear = 0.775

constexpr float cg_to_nose = 1.5f

constexpr float whl_radius = 0.2286

constexpr float gear_ratio = 15.0f

gear ratio = motor speed / wheel speed = wheel torque / motor torque

constexpr float car_mass = 210.0f

constexpr float rolling_drag = 100.0f

constexpr float long_tractive_capability = 2.0f

Maximum forward acceleration in m/s^2. Can be an imposed limit or the actual physics limitation.

Total drag forces on the car (rolling friction + air drag) in N

constexpr float lat_tractive_capability = 3.0f

Maximum centripetal acceleration in m/s^2. Can be an imposed limit or the actual physics limitation. Usually slightly more than long_tractive_capability

constexpr float understeer_slope = 0.0f

How much car understeers as speed increases. See.

Model Overview .

constexpr float brake_enable_speed = 1.0f

constexpr float saturating_motor_torque = (long_tractive_capability + rolling_drag / car_mass) * car_mass * whl_radius / gear_ratio

Maximum torque request (N m)

constexpr float approx_propogation_delay = 0.02f

Time from MPPI control action request to physical change, in sec.

constexpr float approx_mppi_time = 0.02f

Time from MPPI launch to control action calculation, in sec.

constexpr TorqueMode torque_mode = TorqueMode::FWD

Because back two wheels don’t currently work.

constexpr uint8_t state_x_idx = 0

constexpr uint8_t state_y_idx = 1

constexpr uint8_t state_yaw_idx = 2

constexpr uint8_t state_speed_idx = 3

constexpr uint8_t action_swangle_idx = 0

constexpr uint8_t action_torque_idx = 1

constexpr const char *clear_term_sequence = "\033c"