Carla

* Revamped driving benchmark
- Changed name from benchmark to driving benchmark
- Fully Redesigned the architecture of the module
- Added a lot more documentation
- Now you can stop and resume the benchmarks you run
* Rolled back vehicle's location to the pivot of the mesh instead of the center of the bounding box
* Added relative transform of the vehicle's bounding box to the measurements, player and non-players
* Added "frame number" to each sensor measurement so it is possible to sync all the measurements based on the frame they are produced
* Improved vehicle spawner to better handle spawning failures
* Walkers use now a closer angle to detect vehicles, so they don't stop moving if a car passes nearby
* Fixed lighting artefact causing the road to change its brightness depending on the distance to the camera
* Fixed captured images overexposed in Low mode
* Fixed illegal character in asset name
* Fixed editing sun azimuth angle in CarlaWeadther.ini had no effect
* Fixed crash when using a non-standard image size in DirectX (Windows)
* Fixed issue with using multiple "SceneCaptureToDiskCamera"

* New Python example for visualizing the player start positions
* Fixed box extent of non-player agents was sent in centimeters instead of meters
* Fixed speed limits were sent in km/h instead of m/s
* Fixed issue in Volkswagen T2 wheels causing it to overturn

* Upgraded to Unreal Engine 4.18
* Created our own pedestrian 3D models free to use and distribute
* Removed Epic's Automotive Materials dependencies
* 360 Lidars support (similar to Velodyne HDL-32E or VLP-16) thanks to Anton Pechenko (Yandex)
- Ray-cast based
- Configurable settings
- Added methods to save points to disk as PLY file
* Added quality level settings
- Low: low quality graphics, about 3 times faster with one camera
- Epic: best quality (as before)
* Measurements now use SI units
- Locations: m
- Speed: m/s
- Acceleration: m/s^2
- Collisions: kg*m/s
- Angles: degrees
* Added API methods to convert depth images to a point cloud
- New method "image_converter.depth_to_local_point_cloud"
- A supplementary image can be passed to attach colors to the points
- New client example generates a point cloud in world coordinates
- Added Transform class to Python API
* Performance optimizations
- Significant speed improvements in both Epic and Low modes
- Fixed materials and improved shaders for roads, architecture, sidewalks, foliage, landscapes, cars, walkers, reflections, water
- Execution of a set of Project and Engine parameters to improve performance (quality, vsync, AO, occlusion)
- Generation of the road pieces using static meshes and actors instead of a single actor with instanced meshes
- Improved performance since now is able to apply occlusion and draw distance
- Images are captured asynchronously in the render thread
- In asynchronous mode, images may arrive up to two frames later
- In synchronous mode, game thread is blocked until images are ready
- Blueprint code optimizations for vehicles, walkers, and splines
- Added a way to configure different quality levels with culling distance and materials configuration
* Refactored sensor related code to ease adding new sensors in the future
* Added vehicle box extent to player measurements
* Removed the player from the list of non-player agents
* Adjusted bounding boxes to vehicles' height
* Changed vehicles' center to match bounding box
* Added autopilot mode to manual_control.py
* Added quality level options to manual_control.py and client_example.py
* Replaced background landscape and trees by a matte painting
* Fixed road map generated some meshes twice
* Small improvements to Windows support
- Fixed issues with the Makefile
- Fixed asset names too long or containing special characters

* New Python API module: Benchmark
- Defines a set of tasks and conditions to test a certain agent
- Contains a starting benchmark, CoRL2017
- Contains Agent Class: Interface for benchmarking AIs
* New Python API module: Basic Planner (Temporary Hack)
- Provide routes for the agent
- Contains AStar module to find the shortest route
* Other Python API improvements
- Converter class to convert between Unreal world and map units
- Metrics module to summarize benchmark results
* Send vehicle's roll, pitch, and yaw to client (orientation is now deprecated)
* New RoutePlanner class for assigning fixed routes to autopilot (IntersectionEntrance has been removed)
* Create a random engine for each vehicle, which greatly improves repeatability
* Add option to skip content download in Setup.sh
* Few small fixes to the city assets

* New Python client API
- Cleaner and more robust
- Compatible with Python 2 and 3
- Improved exception handling
- Improved examples
- Included methods for parsing the images
- Better documentation
- Protocol: renamed "ai_control" to "autopilot_control"
- Merged testing client
- Added the maps for both cities, the client can now access the car position within the lane
* Make CARLA start without client by default
* Added wind effect to some trees and plants
* Improvements to the existing weather presets
* Build script: skip content download if up-to-date

* Included Unreal project and reorganised folders
* Enabled semantic segmentation by default
* Added Felipe's Python client
* New build system (Linux only)
* Few fixes to city assets