opt_control generates time-optimal second- and third order trajectories from arbitrary start- to arbitrary target states. The trajectories respect per-axis constraints on minimum and maximum velocity, acceleration and jerk. Individual axes can be coupled by synchronizing the total time of each trajectory. Since the method is very fast (<<1ms per axis per trajectory), it can be used in closed loop even for fast systems. With the ability to predict the target state, trajectories end in an optimal interception point when the waypoint is non-stationary. The method has been successfully used as model predictive controller on different micro aerial vehicles, in different research projects and robotic competitions.
Marius Beul and Sven Behnke: Fast Full State Trajectory Generation for Multirotors
In Proceedings of International Conference on Unmanned Aircraft Systems (ICUAS), Miami, FL, USA, June 2017
Marius Beul and Sven Behnke: Analytical Time-optimal Trajectory Generation and Control for Multirotors
In Proceedings of International Conference on Unmanned Aircraft Systems (ICUAS), Arlington, VA, USA, June 2016
@INPROCEEDINGS{beul2017icuas,
author = {Marius Beul and Sven Behnke},
title = {Fast Full State Trajectory Generation for Multirotors},
booktitle = {Proceedings of International Conference on Unmanned Aircraft Systems (ICUAS)},
year = 2017,
month = 6,
address = {Miami, FL, USA},
doi = {10.1109/ICUAS.2017.7991304},
url = {https://github.com/AIS-Bonn/opt_control}
}
run opt_control_example in MATLAB.
The variable index_example selects the example.
- Example 1 generates a predefined trajectory with two consecutive waypoints for a single axis.
- Example 2 generates a predefined trajectory with two consecutive waypoints for a single axis with changing bounds and infeasible start state for the second waypoint.
- Example 3 generates a predefined second order trajectory with two consecutive waypoints for a single axis.
- Example 4 generates a predefined synchronized trajectory with two consecutive waypoints for two axes.
- Example 5 generates a random trajectory with up to five consecutive waypoints for up to ten axes.
Given that the trajectory consists of m axes and n consecutive waypoints, the variables have the following size:
| Variable | Type | Size |
|---|---|---|
| State_start | double | (m x 3) |
| Waypoints | double | (m x 5 x n) |
| V_max | double | (m x n) |
| V_min | double | (m x n) |
| A_max | double | (m x n) |
| A_min | double | (m x n) |
| J_max | double | (m x n) |
| J_min | double | (m x n) |
| A_global | double | (m x 1) |
| b_comp_global | bool | (1 x 1) |
| b_sync_V | bool | (m x n) |
| b_sync_A | bool | (m x n) |
| b_sync_J | bool | (m x n) |
| b_sync_W | bool | (m x n) |
| b_rotate | bool | (1 x n) |
| b_best_solution | bool | (m x n) |
| b_hard_vel_limit | bool | (m x n) |
| b_catch_up | bool | (m x n) |
| solution_in | double | (m x 2 x n) |
| J_setp_struct | struct | (m x 1) |
| solution_out | double | (m x 2 x n) |
| T_waypoints | double | (m x n) |
Start state of the whole trajectory. The state consist of position, velocity and acceleration per axis
List of waypoints the trajectory has to cross. Each waypoint is 5-dimensional, consisting of position, velocity, acceleration, velocity prediction, and acceleration prediction (currently not functional) per axis.
The velocity prediction variable describes the motion of the waypoint. This enables the method to generate interception trajectories. Velocity and acceleration are waypoint-centric. This means, if a waypoint is moving, the allocentric velocity in the waypoint is waypoint velocity + waypoint velocity prediction.
Maximum velocity per axis per waypoint. The limit is per axis. So, the relative velocity difference of a waypoint can exceed this value, if the waypoint is moving. If any limit is set to Inf, the computation of the trajectory speeds up further, so remove any unnecessary bounds.
Minimum velocity per axis per waypoint. The limit is per axis. So, the relative velocity difference of a waypoint can exceed this value, if the waypoint is moving. If any limit is set to Inf, the computation of the trajectory speeds up further, so remove any unnecessary bounds.
Maximum acceleration per axis per waypoint. If any limit is set to Inf, the computation of the trajectory speeds up further, so remove any unnecessary bounds.
Minimum acceleration per axis per waypoint. If any limit is set to Inf, the computation of the trajectory speeds up further, so remove any unnecessary bounds.
Maximum jerk per axis per waypoint. By setting this to Inf, the order of the trajectory can be reduced.
Minimum jerk per axis per waypoint. By setting this to Inf, the order of the trajectory can be reduced.
Global acceleration. This can be for example constant sidewind when controlling an MAV.
Should global acceleration be compensated?
Should the velocity of faster axes be reduced to synchronize with the slowest axis. When b_sync_V, b_sync_A, and b_sync_J are switched on (and b_sync_W switched off), the trajectory with the smallest deviation from the time-optimal trajectory is chosen.
Should the acceleration of faster axes be reduced to synchronize with the slowest axis? When b_sync_V, b_sync_A, and b_sync_J are switched on (and b_sync_W switched off), the trajectory with the smallest deviation from the time-optimal trajectory is chosen.
Should the jerk of faster axes be reduced to synchronize with the slowest axis? When b_sync_V, b_sync_A, and b_sync_J are switched on (and b_sync_W switched off), the trajectory with the smallest deviation from the time-optimal trajectory is chosen. This feature is currently not functional.
Should all axes be synchronized by waiting? This means that faster axes wait before starting or after finishing at the waypoint for the slowest axis. This only works when waypoint velocity and acceleration is zero of either the start- or target waypoint. This setting superceeds b_sync_V and b_sync_A, since it is computationally much cheaper. So when b_sync_W is true, the settings of b_sync_V and b_sync_A are ignored and it is synchronized by waiting. This setting however results in non-straight multidimensional trajectories.
Rotate the axes so that the x-axis points into the direction of movement between waypoints. This only works with more than one axis. Considering a 3-dimensional trajectory (x,y,z), the coordinate system is rotated about the z-axis.
Should the best solution be returned? See solution_in for details.
This only affects trajectories that start in an infeasible (outside of allowed velocity or acceleration margins) state. If the start velocity is infeasible, should we return to the allowed bounds as fast as possible (b_hard_vel_limit == true) resulting in the shortest possible time outside the limits, or is it allowed to stay outside a little longer for shortest total time and a smoother trajectory. See also Example 2.
Should later synchronized axes try to catch up with unsynchronized previous axes? So, should they catch up in time to synchronize again?
Each trajectory connecting two waypoints for a single axis can be one of 24 cases (time-optimal) and one of 28 cases (synchronized). If b_best_solution is set to false, not all cases are tried to find a solution for the problem, but when a solution is found, return. If you know the case beforehand (e.g. from a previous run with similar waypoints), you can provide the initial case here. If the case is not suitable for the problem however, the method will try the other cases until it finds a solution. Set solution_in to -1 if you have no clue about the possible case.
This variable outputs the switching times of the trajectory with the corresponding jerks. This is why we do all the math...!
This encodes the cases of the time-optimal and synchronized trajectories.
This variable gives the incremental times between waypoints for each axis.
opt_control is licensed under the BSD 3-clause license.
Marius Beul <mbeul@ais.uni-bonn.de>
Institute of Computer Science VI
Rheinische Friedrich-Wilhelms-Universität Bonn
Endenicher Allee 19a
53115 Bonn