ADAS (Advanced Driver Assistance Systems) and Active Safety Systems aim at helping the driver during his driver process, releasing him from certain tasks thanks to automated systems and keeping him, his car occupants and the car surroundings in a high safety level.
Some systems are already very common in commercialized cars (ABS, ESP, cruise control, navigation, automated lights switch on/off, …) Others are more complex and require access to multiple heterogeneous sensors and actuators in order to ensure functions such as accurate positioning, environment perception, situation analysis, and take up the necessary actions (automated parking, pre-crash braking, blind spot detection, lane following, adaptive cruise control…)
The development process of such systems requires real time access to sensors as heterogeneous as video cameras, lidars, radars, vehicle CAN bus, GPS, and so on... and an easy way to integrate new functions and algorithms. In such contexts, developers often face challenges (not to say nightmares) such as multi-threaded programming, data samples timestamping and re-synchronization, data latencies measurements and estimation, system optimization and performance assessment, code re-use and software applications maintenance,…
RTMaps provides a way to prototype efficiently on new algorithms and new systems by providing a modular environment to easily test and evaluate functions based on different sets of sensors, in different configurations, and different sets of processing and data fusion strategies.
RTMaps datalogging and real time playback capabilities allow to work offline on real data, in a reproducible manner for sensors and algorithms development and benchmarking.
HIGHLIGHTS
Support for many automotive sensors (CAN & LIN bus, video cameras, stereo vision heads, GPS, inertial sensors, radars, lidars, DAQ…)
Sensors & subsystems benchmarking
Processing and fusion algorithms development, testing, validation and benchmarking (to be easily integrated via the RTMaps C++ SDK)
Data timestamping, latencies measurement, downstream resynchronization
Datalogging / Data playback
APPLICATION EXAMPLES
Development and Testing
1. Datalogging and distributed Datalogging Datalogging capabilities are fundamental to understand the behavior of a vehicle and its embedded systems. Furthermore, RTMaps provides a way to rapidly playback the many recorded data sources in real-time in order to be able to work offline on algorithms development and fine tuning. While playing back sensors datasets in the lab, RTMaps emulates the availability of the real sensors for the downstream components, hence provides reproducible conditions in order to be able to test and compare different algorithm settings or processing techniques, evaluate algorithms behavior in many characteristic scenarios, while preserving the capability to easily deploy the said functions back into a real prototype vehicle.
Distribution capabilities of RTMaps also provide a way to synchronize the clocks of several RTMaps software instances over multiple computers equipping different systems (like in different cars, on infrastructure elements,…). Once RTMaps clocks are synchronized, it is straightforward to perform synchronized sensors data recordings in various places. At playback, it becomes possible to work on a distributed cooperative system offline in just one place !
2. Datafusion for obstacle detection A single sensor usually does not provide enough information to support a function like obstacle detection in a moving vehicle in all conditions (by day, by night, by fog, by rain, …) See the FADE application developed under RTMaps by the CAOR (Center of Robotics, Ecole des Mines de Paris) which associates several monovision based algorithms and a long-range radar to achieve robust detection in many various situations.
Benchmarking & Validation
1. Sensors and algorithms benchmarking Having developed or purchased a perception system or algorithm, whatever it is for (obstacle detection, positioning, road signs detection, etc.), it is always very difficult to assess its performance and compare it to other existing methods.
With RTMaps, it is easy to setup an application which can record the outputs of the system to evaluate in parallel to the outputs of a reference system providing ground truth information (based on video, or expensive sensors like RTK GPS receivers, 3D laser scanners, etc.)
With RTMaps, engineers can record characteristic scenarios operating multiple sensors (by day, by night, by fog, by rain, inside tunnels, etc.) and use them offline to feed perception algorithms in playback mode and evaluate their respective performance (false-detection rate, non-detection rate, …)
CVIS used RTMaps for the development and benchmarking of a high frequency accurate positioning system based on low-cost sensors
Deployment
1. Porting an application to an embedded target (DSP, FPGA, ASIC…) The RTMaps architecture is based on a very powerful and optimized runtime engine, written in C++ and running pre-compiled components also written in C++. This makes it really easy to use (drag & drop, one-click execution, auto-configuration and plug & play devices handling…) and very dynamic (parameterization and diagram edition is possible while an application is executing). Running an RTMaps application hence requires at least a minimalistic OS layer (Windows embedded, micro Linux…). However, systems developed with RTMaps often aim at being ported to embedded targets such as DSPs, FPGAs or so. This is why RTMaps also provides a close integration with Mathworks Simulink®, both in co-execution (or co-simulation) mode for devlepment and testing stages, and via a Simulink® RTW Embedded Coder TLC target for generating C code from a Simulink® model into a runnable RTMaps component. Such a model can then be ported to an embedded target after having been validated in RTMaps.
References
Renault -------------------------------------------------------- PSA Peaugeot Citroen -------------------------------------------------------- Valeo -------------------------------------------------------- Ichikoh -------------------------------------------------------- Ecoles des Mines ParisTech -------------------------------------------------------- LiVIC --------------------------------------------------------
