Introduction
Every robot added to a production line cuts labour hours. But it shifts the cost elsewhere: programming. Today, between 20% and 30% of a robotic cell’s commissioning budget goes into manually working out which robot does which task, in what order and with what trajectory, so they do not collide.
When a cell has a single robot, this work is already demanding. When several robots share the same space, as happens in most of our automotive welding projects, the problem becomes far more complex and requires a great deal of manual expertise to solve efficiently.
That is why we launched RITMO (an intelligent platform for the automatic optimisation of trajectories and tasks in multi-robot industrial cells), an R&D project we are developing between 2025 and 2027 with co-financing from the Institut Valencià de Competitivitat Empresarial (IVACE), the Generalitat Valenciana, and European Funds, under the Business Value Chain Consolidation programme.
The problem the market hasn’t solved
Before deciding to build our own solution, we tested what already exists. We evaluated the most advanced platforms we found on the market for generating multi-robot routes. We even had direct support from the vendor to adapt it to our real cells.
The result: the software works in simple cases, but it did not run reliably in the complex cases we handle, with multiple constraints and manufacturer-specific rules. Other alternatives are built for logistics and palletising, not for assembly trajectories that must meet very specific safety requirements. And the major robot manufacturers (ABB, Siemens, FANUC) include simulation modules, but simultaneously optimising several robots with complex constraints still requires manual programming or specialised external tools.
This gap is not unique to us. Spain installed more than 5,160 new robots in 2024, its second-best year on record, and the global installed base already exceeds 4 million units (World Robotics 2024, IFR). Despite this growth, there is still no software that reliably automates joint planning for multiple industrial robots at the level large industry demands.
Why this is such a hard problem to solve
Coordinating several robots in the same cell is not simply “harder” than coordinating one: it is a different kind of problem. Each weld point can be reached through several possible joint configurations. Each robot can visit its assigned points in different orders. And every combination has to be checked to rule out collisions between robots, with the tooling, or with the part.
Technically, this is a variant of the Multiple Traveling Salesman Problem (mTSP), an NP-hard combinatorial optimisation problem. Exact solution techniques (dynamic programming, mixed-integer linear programming) become intractable even with a small number of weld points, which means combining heuristics, hierarchical planning, and search-space reduction techniques.
We do not tackle this challenge alone. We collaborate with two teams from the Universitat Politècnica de València: the Institute for Design and Manufacturing (IDF), which brings expertise in robotics and motion planning, and the Department of Systems Engineering and Control (DISA), specialised in control, process optimisation and multi-agent systems. Together they bring key capabilities for planning multi-robot trajectories and optimising their execution in complex industrial environments.
What we are building
RITMO is organised into four technology blocks:
Optimisation engine. Selects the best joint configuration for each work point, plans collision-free routes between pairs of points, and solves the optimal visiting sequence, first for a single robot and then extending it to several robots working in parallel.
AI module. Generates synthetic trajectory datasets and trains surrogate models and reinforcement-learning agents able to predict performance metrics in seconds and to correct deviations between simulated and real robot behaviour.
CAD/PLM data adapters. Natively connect our current workflow (Siemens Process Simulate) with NVIDIA Omniverse, the simulation and digital-twin platform increasingly adopted by the industry, ensuring geometry, weld-point and metadata transfer without friction between both environments.
Graphical interface in Omniverse. Unifies cell configuration, the launching of the different optimisation engines, and results visualisation, designed both for our own team and so that in the future we can also put it at the service of other teams.
Why we use AI, and for exactly what
RITMO’s AI isn’t a superficial addition: it addresses a very specific computing constraint. Simulating the behaviour of several robots under real loads with precision is a costly process, and a full simulation run in an environment such as CoppeliaSim or Omniverse can take minutes. When the optimisation engine needs to evaluate thousands of possible trajectory combinations to find the best one, that time becomes a bottleneck in itself.
That is why we train surrogate models: neural networks that learn, from a synthetic dataset generated with the simulator, to predict in seconds the same metrics that previously required a full simulation (execution time, joint effort, collision risk). This lets us explore a much wider solution space in the same amount of compute time.
The second use of AI is equally concrete: reinforcement-learning agents correct trajectories when the robot’s real behaviour, subject to real loads and the real mechanics of the cell, drifts from what the simulation predicted. It is the difference between a trajectory that works in the simulator and one that works on the shop floor.
In both cases, AI does not replace the optimisation algorithms or the engineering knowledge we already have: it speeds up specific parts of the process that, calculated with traditional methods, would be too slow for real industrial use.
What this means for RBX and for the industry
This project marks a shift in how we will be able to help our clients. Today we support them mainly with specialised engineering in virtual commissioning, robot programming, PLC programming and electrical engineering, project by project. With RITMO, we will also be able to help them with technology of our own that drastically reduces manual programming hours, development time, and cycle time on the shop floor.
This lets us reach more clients with different needs, keep strengthening the support we already offer compared to those who only work with manual programming, and also help companies in sectors where we have less presence today, such as logistics or food, alongside our current automotive focus.
For the rest of the value chain, the expected effect is just as relevant. By reducing the cost and complexity of programming multi-robot cells, RITMO can help other, smaller integrators adopt advanced robotics without needing to develop this technology themselves, and help OEM manufacturers receive cells with shorter cycles and lower energy consumption. We have already seen this in practice in our Virtual Commissioning work: in this case study with one of the largest EV manufacturers in the US, reducing commissioning complexity translated into shorter cycle time and fewer engineering hours.
Next steps
RITMO has a three-year development horizon (2025-2027). We will keep sharing progress as we complete each phase, from the first results of the single-robot optimisation engine to final validation on real pilot cells.
If you work with multi-robot robotic cells and want to know more about the project, or explore how it could apply to your operation, get in touch.