New benchmarks show how RoboDK 6 supports responsive path planning, runtime adjustments and emerging physical AI applications.
Collision checking is one of the most computationally demanding parts of robot simulation and motion planning. A planner may need to check thousands of potential robot configurations and path segments against the surrounding geometry. That’s a lot of mathematics.
Those calculations are now significantly faster in RoboDK 6. In fact, ‘significantly’ may be an understatement. Depending on the application and hardware configuration, RoboDK 6 performs collision checking up to 100 times faster than previous versions.
This level of performance opens up new opportunities for responsive path planning, runtime adjustments and emerging physical AI applications.
Using a complex automotive spot-welding station containing a Comau Smart5 NJ 130-2.6 robot, RoboDK recorded collision-checking rates of up to 4,357 samples per second. In a second test, RoboDK checked all 3,589 steps of the robot program at up to 1,204 steps per second—an average of just 0.83 milliseconds per step.
Tests across five hardware and software configurations recorded full-program collision-checking rates ranging from 83 to 1,204 steps per second. Results will vary depending on the processor, operating system, RoboDK build, station complexity and collision settings, but overall, collision checking can now be performed much more quickly across complete robot programs and multiple hardware configurations. So, what does all this mean?
The value of faster collision checking
Traditional collision-free motion planning algorithms use probabilistic roadmaps. Software samples possible robot configurations, identifies collision-free positions and connects them to create a navigable map through the robot’s workspace. When the robot needs to move between two positions, the software tries to quickly connect and navigate through this ‘map’ for a valid route. Having this map helps speed up calculation of collision-free paths as traditional hardware and software integration used to be slow.
This works well while the environment remains substantially unchanged. But when a vehicle body, workpiece or other object starts moving, the map may not match the sampled territory, and a previously valid path may now contain an obstruction. The system then has to check the updated environment and find an appropriate response quickly enough for the result to remain useful.
RoboDK 6 enables applications to perform these calculations much faster. To the point where sampling the workspace for a probabilistic roadmap is not needed anymore. This in turn enables RoboDK users to:
- incorporate collision checking with real time control algorithms
- respond to updated object positions, and
- remove the need to precalculate a probabilistic roadmap.
The end result? Faster workflows and more responsive applications.
Consider a robot working on vehicle bodies that are moving on an assembly line. The system can accommodate small variations in the positioning of the car body by quickly updating the digital twin using RoboDK and checking the newly adjusted path -instead of relying entirely on the route calculated for the original position.
Another immediate benefit is faster simulation and program validation. Manufacturers and integrators can identify collisions, refine robot paths and validate complex programs in less time.
For example, programs for automotive spot welding can contain thousands of closely spaced movements around the vehicle structure, welding guns and fixtures. Faster checking enables people deploying automation to evaluate path changes and validate the complete program much faster.
Supporting physical AI
AI systems use cameras and other sensors to interpret scenes, identify targets and propose robot actions. But before that action is executed, the proposed robot movement must be evaluated against the physical constraints of the application and the environment.
In vision-guided bin picking, for example, an AI system may identify a different object and generate a new grasp on every cycle. The robot then needs to determine whether it can approach, grasp and remove that object without striking the bin, neighbouring parts or surrounding equipment.
RoboDK provides the simulation and robot-programming layer within such processes. An external perception or AI system can update the digital twin through the RoboDK API. RoboDK can then check the proposed motion for collisions and support adjustment or replanning before the validated movement is transferred to the physical robot.
Faster collision checking shortens the loop between perception, planning and action. It moves collision evaluation beyond purely offline verification and toward a more active role in responsive robot applications.
Consider a mobile manipulator moving between workstations. It may encounter carts, pallets or equipment that are not in their previous, expected positions. Rapid collision checking can help its planning system evaluate alternative motions when the surrounding workspace changes.
RoboDK 6’s collision checking speeds also provide developers with a fast and effective way to test whether their AI-generated motions are geometrically valid.
RoboDK has published the benchmark application, test station, complete results and command-line procedure on GitHub. Developers can inspect the methodology and run the tests on their own systems.
Explore the RoboDK 6 performance results and benchmark.
Join the discussion on LinkedIn, Twitter, Facebook, Instagram, or in the RoboDK Forum.. Also, check out our extensive video collection and subscribe to the RoboDK YouTube Channel.
