At a glance: Sorbonne’s ASIMOV lab builds robots for unpredictable real-world environments, like GNSS-denied drainage tunnels and soft snake-like manipulators, where precise spatial awareness is the core challenge that makes everything else possible.
Dr. Ouarti Nizar doesn’t build robots that work in clean rooms with taped lines on the floor. His team at Sorbonne University’s ASIMOV lab builds robots that crawl through Singapore’s drainage systems hunting mosquito larvae, soft-bodied manipulators that move like snakes, and swarms of tiny machines that coordinate through light signals.
These aren’t the robots you see in controlled factory environments. They’re designed for the messy, unpredictable places where precision matters most.
“The arm can’t move the same way as a human arm,” Dr. Ouarti explains when discussing the constraints his team navigates daily. It’s a simple statement that captures a complex truth: teaching machines to interact with the real world means accepting that they’ll never be exact replicas of biological systems. Instead, his lab focuses on making robots that are autonomously intelligent enough to figure things out on their own.
The Drain Bot Problem
Take the drainage inspection robot developed for Singapore’s public health infrastructure. The mission sounds straightforward: detect stagnant water where mosquitoes breed, identify cracks in tunnel systems, and create 3D maps of underground infrastructure. But getting there requires solving multiple problems simultaneously.
The robot needs to know where it is in a dark, GPS-denied tunnel. It needs computer vision sophisticated enough to distinguish between flowing water and stagnant pools. It needs to build spatial models accurate enough that maintenance teams can pinpoint exactly where to deploy larvicide and not flood entire systems with chemicals, but target the specific pockets where mosquito eggs concentrate.
This means precision location isn’t a nice-to-have feature here. It makes the difference between effective disease prevention and wasted resources.
Soft Robotics, Hard Problems
The lab’s soft robotics work presents even stranger challenges. Traditional rigid robots operate in known geometric spaces: joint angles, end-effector positions, and predefined movement planes. Cable-driven soft robots that move like biological organisms? The math gets complicated fast.
“If you change something, the shape doesn’t maintain,” Dr. Ouarti notes about their snake-like manipulator. The robot’s configuration depends entirely on its starting position. Every movement creates a new geometric relationship that the control system needs to understand in real-time.
Making this work requires fusing multiple sensor streams, computer vision tracking the robot’s continuously changing shape, motor encoders, force feedback, all into a coherent understanding of where every part of the system exists in space at any given moment.
Teaching Robots to Grasp Reality
The lab tackles problems most people never think about. Where exactly do you grip a cup? Handle or rim? How does a robot arm working with L’Oréal learn to recreate identical shapes across thousands of repetitions? How do you program a three-dimensional trunk-like appendage when its movement is organic rather than mechanical?
These questions all bottom out in the same place: robots need to understand their position in space with enough precision to interact meaningfully with objects and environments that don’t care about their limitations.
The ASIMOV team runs the full spectrum, from collaborative art installations with enormous robotic arms to swarm robotics small enough to fit in your palm, from autonomous cars using LiDAR to hospital mobility units navigating hallways filled with unpredictable humans.
What connects all of it is the fundamental challenge of autonomous operation in “open environments”, the lab’s term for anywhere that isn’t perfectly controlled and mapped in advance.
The PhD Students Building Tomorrow
Most of the 37 researchers in the lab are PhD students or postdocs. Instead of just developing software, they’re building custom hardware, teaching 9-year-olds to program mobile robots through block-based interfaces, and pushing into territories where simulation fails to capture reality’s full complexity.
The “simulation-to-reality gap” shows up constantly in their work. Algorithms that work perfectly in virtual environments stumble when deployed in actual drains, actual manipulation tasks, actual environments where lighting changes, surfaces are uneven, and unexpected obstacles appear.
Bridging that gap requires robots that can perceive their environment accurately, process sensor data in real-time, and make autonomous decisions when the programmed response doesn’t quite fit the situation they’ve encountered.
The ASIMOV lab isn’t happy with only incremental improvements to existing systems. This is fundamental research into how machines achieve autonomous intelligence in conditions where precision location, real-time perception, and adaptive decision-making determine whether the robot completes its mission or becomes an expensive paperweight.
The robots coming out of labs like ASIMOV won’t just work ,they’ll work in conditions where GPS doesn’t reach, where the environment changes constantly, and where centimeter-level accuracy separates success from failure.
That’s when knowing exactly where you are stops being a technical specification and starts being the foundation for everything else.