Hidden Challenges of Deploying Robots in the Real World
Part 1: Mapping & Route Setting
Robots rely on maps to navigate their environments. But anyone who's ever walked through a hospital knows that “map” is a generous term for what’s essentially a chaotic, ever-changing obstacle course. What looks like a static floorplan on paper is, in reality, a living, breathing space filled with rolling beds, surprise furniture, and ceiling-mounted contraptions that seem specifically designed to confuse sensors.
In this post, I’ll dig into the behind-the-scenes work we’ve done at Akara to help our robots find their way through this complexity. From keeping maps accurate in dynamic hospital environments, to building custom navigation tools for disinfection tasks, to making sure multiple robots share the same understanding of the world, here are some of the things that it takes to get robots to operate reliably in the real world.
Keeping maps up-to-date
Hospital layouts are not static — rooms are reorganized, new areas are added, and furniture is frequently moved. We addressed this by:
Working closely with frontline hospital staff to create a user-friendly interface that allows non-robotics experts to create and update maps themselves.
Implementing functionality that allowed map-building through remote teleoperation. This enabled us to create new maps without being physically on-site.
Handling obstacles not on the map
Most robot maps are essentially 2D floorplans that lack details such as movable furniture. To develop robustness so that the robot’s behavior can adapt to these uncertainties, we found it necessary to:
Create obstacle avoidance software capable of engaging in custom fallback behaviors if its path becomes unexpectedly blocked or it encounters unexpected objects not represented on the map.
Integrate a network of depth sensors mounted on the robot to detect elevated obstacles, since we realized that hospitals often have ceiling mounted instruments that are often not detected by the standard sensor suite found on most mobile robots.
Sharing maps across a fleet of robots
When deploying multiple robots, ensuring they operate with consistent maps is crucial. To solve this problem, we implemented a bespoke version control system centralized storage of maps and routes. Maps are stored in a remote server and robots automatically load these maps during startup. This ensured all robots had the most up-to-date maps and that outdated maps were phased out systematically.
The challenges I’ve discussed so far are fairly universal as they affect almost any autonomous mobile robot operating in the real world. But developing robots for UV disinfection introduces a new layer of complexity. Unlike typical navigation tasks, disinfection requires careful planning to ensure full surface coverage, avoid shadows, and deliver the right dose of UV light. The next sections dive into the unique technical hurdles we faced and how we tackled them.
Route setting for UV disinfection
Setting effective routes for our UV robots posed unique challenges. Like robotic vacuum cleaners, UV robots are required to travel to many locations to ensure the maximum surface coverage of the germ-killing UV light. This requirement is distinct from the majority of mobile robot applications that focus on point-to-point navigation, where the primary objective is to travel from one location to another as quickly as possible, following the simplest route. This fundamental difference in goals meant that in our application, even the state-of-the-art navigation algorithms were not suitable without significant customization.
To make it possible for our robot to reliably travel to many waypoints in a room and navigate tight, cluttered regions that might be inaccessible to other mobile robots, the following actions were taken:
Industry-leading expertise: We hired one of the world’s leading experts on robotic navigation. With their guidance, we were able to customize a motion planner optimized for our specific use case.
Virtual robotic simulation: Significant efforts were invested in the development of simulation tools that enabled different robot navigation algorithms to be empirically evaluated in virtual environment that emulated the kinds of hospitals our robots were required to operate in.
User tools: Developed software tools to simply the process of setting waypoints on the map, improving usability.
Minimizing shadowed areas
UV light only disinfects surfaces that are in direct line of sight of the UV lamps. Therefore, during a UV cleaning procedure it can be hard to know which surfaces were exposed to the UV light and were decontaminated, and which ones were not. This motivated us to develop what I believe is the first scalable method that assures optimal surface coverage. It involves:
Creation of digital-twin: We created tools to create high fidelity 3D maps of the hospitals where our robots would be deployed. These 3D maps subsequentially formed the basis for a digital-twin used in simulation.
Light modelling: We used ray-tracing techniques to develop detailed models of the UV irradiance field generated by our robot. By integrating these light models into our simulation environment, we were able to accurately visualize how UV light interacted with the surrounding space. This allowed us to identify which surfaces received exposure during a disinfection procedure and, just as importantly, which areas remained in shadow and did not receive UV radiation.
Route creation protocols: Using the above tools, protocols were developed to optimize the positioning of the robot to ensure the maximum number of surfaces were irradiated.
Ensuring sufficient UV dose
For UV light to neutralize pathogens effectively, a critical dose of energy must be delivered. Determining how long a robot needs to remain in position to achieve this dose is not trivial because UV energy decreases non-linearly with distance. To solve this we created:
Parametric UV dose model: We developed and validated a parametric physics model for quantifying the light energy produced by our lamps, allowing us to accurately estimate the UV dose administered to surfaces surrounding the robot accounting for variable such as lamp size, room reflectivity, and environmental factors. This work culminated in a research paper published in Biomedical Physics and Engineering Express (available here), providing a robust framework for ensuring effective disinfection.
Simulation plugin: We created a software plugin to implement this physics model in our simulations. This enabled us to quantify the dose that would likely be delivered to surfaces and led to further optimization of how the robot was deployed.
Summary
Mapping and route setting might sound straightforward, but in real-world deployments, they present a myriad of challenges that require custom solutions. From managing dynamic environments to ensuring comprehensive disinfection coverage, we had to rethink traditional approaches to robotic navigation and develop a number of specific tools tailored to our use case.
These efforts have not only allowed us to deploy robots that operate autonomously and reliably but have also set a strong foundation for scaling our systems across multiple sites. By addressing these challenges, we’ve gained valuable insights into what it takes to make robots truly work in the real world.
In the next post, I’ll discuss how we tackle safety challenges to ensure that robots can operate securely in sensitive environments like hospitals.