Sub-system
Simulation serves two very different customers in robotics. An integrator uses it to design and program a cell before it is built. A machine-learning team uses it to generate experience faster than reality. Both need the model to match the real thing, and that matching is the hard part.
In one sentence
Robotics simulation models a robot's kinematics, its tooling and its environment — and, for learning applications, the physics of contact — so that programs can be developed and policies trained without the physical system.
The engineering use is mature. A model of the cell verifies reach and collisions, generates motion paths from part geometry, estimates cycle time and lets a program be written while the cell is still being built. Its value depends entirely on the model matching the installed reality, which is why calibrating the simulated robot against the real one is the step that decides whether offline programming works.
The learning use is newer and much more demanding, because it needs the physics to be right rather than the geometry. Rigid-body motion simulates well; contact — friction, deformation, the exact moment surfaces slip — does not. That gap is why locomotion policies transfer from simulation far more readily than manipulation policies do.
Running thousands of simulated robots in parallel on accelerators, rather than one at a time on a processor, turned reinforcement learning for robots from a research curiosity into a practical method. It is the same hardware and the same parallelism as model training, applied to generating experience instead of consuming it.
Since no simulator matches reality exactly, the standard technique is to randomise the parameters a policy might over-fit to — friction, mass, sensor noise, lighting — so it cannot depend on any exact value. The policy learns something robust rather than something tuned to a simulator, which is what allows it to work on hardware.
Technology dependencies are solved by engineering; supply dependencies are solved by building something, which takes years.
Running many simulated environments in parallel is an accelerator workload, and it is what makes the method practical.
AI computeA model that does not match the installed robot produces programs that collide; calibration is what makes offline work usable.
Sensors and encodersParallel physics solvers are written as accelerator kernels; the throughput that makes simulated training practical comes from that layer rather than from the physics.
Programming modelsOther pages in this map that name Simulation and digital twins as something they cannot do without.
What each company supplies at this step, and — where a public figure exists — its share of this specific market — with what that share measures, the period it covers and who published it. Some rows also show the company’s own reported revenue for the segment covering this step, which is a different thing: it says how much this business matters to that company, not how much of the market it holds. Not a ranking and not a recommendation.
Supplies accelerated simulation platforms for robot design and learning.
Supplies factory and robot simulation within its industrial software portfolio.
Supplies manufacturing simulation and digital twin software.
Supplies engineering simulation software following its acquisition of Ansys.
Supplies the design and service-side twin, which is the half that runs after the machine ships.
Supplies the measurement and simulation stack, and the metrology that tells you the twin is still true.
Supplies the design and factory-layout tools most cells are laid out in before anything is simulated.
Supplies the control-design and simulation environment most robot control laws are prototyped in.
Each page explains one technology in plain language, states what it depends on, and names companies by what they supply at that step. Company roles are described qualitatively and deliberately carry no market shares, revenue figures or rankings — those change faster than an explainer can, and a stale number is worse than none. Ticker links point at company pages on this site and are provided for reference only.
Nothing here is investment advice, a recommendation, or a forecast. A company named on a page about a technology is not thereby a good investment, and the chokepoints described are structural facts about supply chains rather than predictions about prices. Technology moves; where a page describes something as unresolved or in development, that was true when it was written.