Integrated Solutions

SCANeR Studio is a well-established integrated solution on the market. From the variety of modules, examples etc. you clearly see that it encompasses many years of experience, many deployments, and the coverage of a large number of use cases. It is also clear that interactive driver-in-the-loop (DiL) simulation is at the roots of this product and left its mark on the real-time capabilities.

Overall, the product can be considered complete in terms of its functionality. This bonus tends to become somewhat of a burden if not all features make it into the integrated GUI and if the number of modules and their inter-dependencies grow continuously.

For users who want to get started and create their first scenarios, the product’s GUI is a big help. The line between SCENARIO and SIMULATION mode, though, seems a bit blurred. For actual deployment of an installation, it appears that some expertise is needed to get everything configured correctly and to get modules operate as desired (but nobody claims that this is, in itself, an easy task at all).

The traffic module provides a good means to create situations that encompass participants of various types. Passenger and commercial vehicles with and without trailers are available as well as various pedestrians and animals. All entities move in quite intelligent fashion within the road network. More complex behavior or dedicated actions may be scripted using either a Python interface or the custom MICE script language which comes with an editing means within the GUI.

Parametric scenarios and cloud deployment provide a good basis for the exploration of event spaces. The lack of a higher-level scenario language (see, for example, the upcoming ASAM OpenSCENARIO 2.0 standard) and of a full orchestration of cloud operation leaves some room for productivity gain in the future.

The sensors of the base product are good in terms of camera features and somewhat more basic in terms of LiDAR and RADAR implementation. With various add-ons (available only to licensed parties), the user may upgrade to fully physics-based sensor simulation which also takes material properties into account and provides a high level of granularity (note: our actual tests were executed on the sensor models of the base product; the features of the physics-based sensors have been cross-checked in a series of interviews with the product management). Overall, the coverage of sensor types (perfect, camera, LiDAR, RADAR, infrared, ultrasonic, map, GPS) looks rather good.

The URenderer (in beta state at the time of our tests), based on Unreal, will be a welcome extension of the product and will definitely improve the standing of the camera-based sensor capabilities in the core product. On the other hand, it creates a dependency to a 3rd party component.

There is, as with all other tools, definitely room for improvement; but with a rating of 3.6 at the time of publication (November 2021), SCANeR Studio is the best rated “integrated solution” yet in our database.

This result is also reflected by the fact that all relevant use cases may be addressed with the tool:

• algorithm development for ADAS/AD controllers
• simulation for planning and control (object lists)
• driver- and hardware-in-the-loop testing under real-time conditions
• verification and validation of ADAS/AD systems
• creation of training data sets for perception systems
• realistic sensor raw data generation and injection across all wavelengths

Features that have not been rated

Our rating system does not claim to be 100 percent complete, and we always try to find a common denominator for tools of similar product category. Tools may, however, provide more features than we are currently rating and comparing (we may include them in future version of our criteria catalogue).

• import and application of IES light distributions for the simulation of realistic headlights
• light sensor for the determination of incident light at a given point in the scene
• cockpit / instrument simulation

This is a nice software to get started with exploring the space of environment simulation for ADAS/AD. Once you get past the basics and need really good and well synchronized simulation of vehicle dynamics, traffic and sensors, you might consider exchanging several components of the simulator (which is feasible via plug-in interfaces) or migrating to a different solution. This solution is developing quite dynamically, therefore checking for progress every now and then seems to be a good idea. This package seems good for Software- and Model-in-the-Loop deployments that do not need full determinism or real-time capabilities.
It's a good fit for the following use cases in ADAS/AD development:

• algorithm development for controllers and basic sensor fusion
• sensor setup and configuration (extrinsic/intrinsic parameters of sensors)
• creation of basic training sets for camera-based ML systems

It has limited value for the following use cases:

• realistic sensor raw data generation and injection across all wavelengths
• hardware-in-the-loop testing under real-time conditions
• verification and validation of ADAS/AD systems

CARLA is the tool of choice for many people who enter the space of simulation for ADAS/AD. It is relatively easy to setup and provides an easy means to integrate and control via Python scripts. Unfortunately, it lacks a GUI for configuration and operation that would make it a much more valuable tool with good usability. CARLA provides some level of determinism but this seems to be limited to the Traffic Manager. The renderer and its physics effects do not appear to provide full determinism. Support is arranged in GitHub but the times for new issues being appointed to someone or being taken care of seem to be too long in many cases.

CARLA seems to be a good fit for the following ADAS/AD development use cases:

• algorithm development for controllers and basic sensor fusion
• simulation for planning and control (object lists)
• sensor setup and configuration (extrinsic/intrinsic parameters of sensors)
• creation of basic training sets for camera-based ML systems

It has limited value for the following use cases:

• realistic sensor raw data generation and injection across all wavelengths
• hardware-in-the-loop testing under real-time conditions
• verification and validation of ADAS/AD systems