Watch a four-module robot walk toward a barrier. Then cut the middle one's throat — all power gone, all sensing gone, all wireless communication gone — and keep watching. It keeps walking.
That is the finding at the center of a paper published February 11, 2026 in Science Robotics by a team at the École Polytechnique Fédérale de Lausanne (EPFL). The robot is called Mori3. The concept is called hyper-redundancy. And for the first time, the researchers say, a modular robot reverses the usual rule: more modules should mean more ways to fail. Instead, Mori3 gets more reliable as it grows — because every module shares everything with every other module.
"We found that sharing just one or two resources was not enough," said Kevin Holdcroft, the paper's first author and a researcher in Jamie Paik's Reconfigurable Robotics Laboratory (RRL) at EPFL's School of Engineering, in the lab's press release. "If each resource had an equal chance of failure, system reliability would continue to drop with an increasing number of agents. But when all resources were shared, this trend was reversed."
The setup matters. Traditional modular robots are fragile in a specific way: one module fails, and the failure cascades. A dead module can't contribute its power, can't feed its sensors into the collective, can't communicate. Depending on the architecture, some or all of the robot's capabilities go with it. More modules, in this model, means more individual failure points — an engineering problem that has kept large swarms and long-duration deployments of modular robots theoretically attractive and practically limited.
Mori3's answer is to make every module a host for every resource. Each triangular module carries power storage, sensing capability, and wireless communication hardware. When a module is cut off from its neighbors, the neighbors flood it with what it needs. In the paper's key demonstration, the researchers severed all three connections to the central module — simulating a complete failure — and watched the robot complete a locomotion task and contort to pass under a barrier using only what neighboring modules contributed.
The revival moment is what Holdcroft called the paper's most striking result. "Our methodology allowed us to revive a dead module in a collective and bring it back to full functionality," he said. The dead module had no power, no sensing, no communication of its own. The neighbors provided all three simultaneously.
The biological parallels are deliberate. The researchers cite birds sharing local sensing through flocking behavior, trees communicating threats to neighbors using airborne signals, and cells continuously transporting nutrients across their membranes so that the death of any individual cell doesn't significantly impact the organism. The paper doesn't oversell the analogy — it is engineering, not biology — but the inspiration is explicit. Mori3 is designed for environments where repair is not an option: space, deep-sea, disaster zones.
"Our aim with Mori3 is to create a modular, origami-like robot that can be assembled and disassembled at will depending on the environment and task at hand," Jamie Paik, head of RRL, said in a separate EPFL feature. The robot's triangular modules can reconfigure into different geometries depending on what a task requires — a shape that fits through a gap, a shape that bridges a distance.
The paper has a clear limitation that the authors acknowledge: every voice in it is from the EPFL team. There are no independent robotics researchers quoted, no external validation of the results, no confirmation from a lab that tried to replicate the finding. The demo involves four modules. Scalability — what happens with 20 modules, or 200, or a swarm of heterogeneous agents — is explicitly listed as future work. The hardware adaptations required for docking energy and data transfer between modules at scale are not yet built.
Those are real gaps, and they're worth naming. A resilience result demonstrated on a four-module robot in a lab is not the same as a resilience result demonstrated on a 50-module swarm in a relevant environment. The paper makes a conceptual contribution that the field will need to test, replicate, and extend.
What the paper does establish is a new structural principle for modular robot design: the three resources — power, communication, and sensing — have to move together. Sharing any two is not enough to reverse the failure-with-scale trend. The insight is clean, the demo is vivid, and the biological grounding gives it intuitive weight that a purely simulation-based result would lack.
The next question is whether the principle holds when the modules are doing something more complicated than walking through a gap — and who builds the first 20-module version that tests it.