px4_config: use directly radians (PX4#500)

Plan to sensor horizon at all times (PX4#374)

This commit removes the 'max_tree_length' parameter, replacing its
logic with the box_size. This way we always plan to the maximum
of the sensor range, which is a logical thing to do.

To throttle the planning effort the number of nodes expanded during
A* search is much more suitable, and that logic remains untouched.

Also Remove turning cost to avoid detours

Merge latest master to stable (PX4#297)

Major changes:

    Visualizations of cost functions
    Execution speed improvements
    Better Realsense clipping radius tuning
    More accurate 'history' of seen data, allowing longer memory and better planning


All changes:

* Cost matrix image (PX4#250)

* add visualization of cost matrix. Fix bug in smoothing function. Take out histogram relevance. set box size to 15m

* lower box size a bit again to 10m

* fix test

* set box size to original value

* Further optimize the smoothing, and add a test

* clean up yaw angle situation a bit. Distinguish between angles in fcu frame and angles in histogram convention (90 deg shifted)

* make smoothing radius a parameter

* clean up

* get rid of unnecessary ROS messaged outside the node

* get rid of hardcoded number

* Make global planner depend on mavros_extras

* README: update how to check if planner is running correctly. (PX4#257)

* README: update how to check if planner is running correctly.
Remove out of date info on multi camera setup

* README:
- rearrange such that system variables don't get overwritten
- fix typi
- MPC_OBS_AVOID -> COM_OBS_AVOID

* Move main processes into subclass functions (PX4#260)

* Make planner info update internal function

* Make publish system status internal function of node

* Fix code style

* Quaternion angle cleanup (PX4#254)

* get rid of tf in quaternion to angle conversion

* fix calculateFOV function, there was a major mixup between frames and rad/deg

* change variable names to have unit in the end and cleanup

* clean up

* take out resolved todo's

* Visualization outfactor (PX4#255)

* factor out rviz visualizations in different class

fix rebase

fix typos and spaces

pass the planner as a const ref. Make all function in planner const which do not change the planner

avoid copying the pointcloud for visualization

add additional diagnostics: pointcloud size as a lightweight topic for logging constraines platforms

make visualization functions const

change location of reference symbols

* cleanup

* pull out drone and world visualization in another class

* disable world loader on hardware

* Don't calculate distance twice and init camera received (PX4#261)

* Initialize node handle in main function (PX4#215)

* Initialize node handle in main

* Fix coding style

* Fix planner node test

* Fix valgridn suppression

* Trajectory simulation (PX4#268)

* Add simple trajectory simulator which goes towards a set velocity vector, limited by jerk and acceleration

* Use higher jerk, and clamp to get accel limits

* Factor out some math, and clamp timestep instead of jerk

* Re-organize into class to share the initial setup

* clang-format

* launch px4 with vehicle model iris_obs_avoid such that COM_OBS_AVOID (PX4#258)

is enabled

* use reprojection of combined histogram for starplanner instead of cropped cloud (PX4#278)

* use reprojection of combined histogram for starplanner instead of cropped cloud

* fix test

* Increase box size due to revised Realsense capabilities (PX4#280)

* fix bug (PX4#282)

use current position in reprojection, not previous position

* Imporve cloud points transformation. (PX4#279)

Transforming cloudpoint consume too much time,especially for multi-cameras,
so create threads to transform cloudpoints parallel.

* Add thesis to docs and readme (PX4#287)

* Add thesis to docs and readme

This commit adds the thesis PDFs by Tanja and Villi explaining the details
of the algorithms.

* Move PDFs to google drive

* Remove legacy code (PX4#284)

* remove stop in front

* remove planner costmap state since it's equivalent to running VFH* with
n_expanded_nodes_:=1

* remove goBack state

* small refactor of determineStrategy() due to removing backOff, stopInFront
and use_VFH_star

* clang format

* remove unused variables

* local_planner: remove obstacle_ bool since hist_is_hempty_ contains the
same information

* Change recursion strategy (PX4#286)

* change recursion strategy

* tune params and discard old points if ouside box

* fix tests

* remove age layer from histogram class

* clean up

* use actual age in seconds to expire data points

* fix tests

* Fix auto takeoff (PX4#285)

* local_planner_node: enable waypoint generator in auto.takeoff and
auto.land otherwise the vehicle is stuck on the ground when
performing the operation from the qgc button

* add return to ground station mode

* remove mission_ and offboard_ variables from local_planner since unused

* add state auto rtl

* path_handler_node: send companion process status active to allow arm (PX4#298)

Test failsafe (PX4#244)

* move avoidance healthiness check from local_planner_node_main
to local_planner_node

* test failsafe

* travis: run roscore in the background

* cleanup test_local_planner_node plus remove active state from
checkFailsafe since it's duplicated

* rename pointcloud_timeout_hover-> timeout_critical and
pointcloud_timeout_land -> timeout_termination

* clang format

* remove cast to float for timeouts since ros::duration requires double

* create another test binary for LocalPlannerNode tests such that
roscore is running only the LocalPlannerNode tests

* don't spin Transfrom Listener thread for tests

* clang format

* default tf_spin_trhread in LocalPlannerNode constructor, fix travis
status variable, fix intendation in CMakeLists

* ROS does not clean up the LogAppender when shutting down, leaving a dangling thread.

The allocation for this thread's stack is caught by Valgrind, so we need to suppress
the error.