By Nicolas Lehment, senior principal system architect, NXP Semiconductors
Automated guided vehicles (AGVs) and wheeled mobile robots currently dominate factories and warehouses, following magnetic stripes or optical markers permanently embedded in warehouse flooring to navigate routes.
As requirements evolve, however, wheels are simply not enough. Demand is growing for machines capable of navigating more challenging, human-centric environments, such as hospitals, restaurants, homes, and even rugged outdoor terrain.
In such environments, perfectly flat floors, free from steps and stairs, along with conspicuous, clutter-free aisles, are luxuries.
Instead, robots must move over thresholds, skirt around unpredictable obstacles and adapt on the fly to a world that’s neither uniform nor pre-mapped.
Humanoid robots are the natural solution, literally following in our footsteps, and there are three foundational areas shaping legged and humanoid robots for such complex settings.
These comprise motion control, perception/navigation, and modularity/flexibility. Together, the three are driving adoption beyond structured industrial floors toward truly versatile autonomous systems.
Controlling motion in unstructured spaces
Traditional factory robots execute pre-planned trajectories over known workspaces – think of three-axis arms that can move in three independent linear directions, for example, or gantry systems moving along fixed rails.
In contrast, legged and humanoid platforms demand real-time, closed-loop control across dozens of degrees of freedom.
Each footfall or joint adjustment must balance stability, torque, and body pose within milliseconds.
Modern embodied robots break the old, centralized motor-controller paradigm. Each joint or limb houses a microcontroller responsible for low-latency torque and position loops, while a central processor coordinates full-body motion plans.
This splitting of duties reduces communication delays and enables smoother, more robust responses to disturbances – crucial when a robot must, for example, steady itself after bumping into an object.
Coordinated multi-motor actions rely on real-time field-bus communications protocols such as EtherCAT, which guarantee sub-millisecond synchronization across dozens of actuators.
Emerging standards include OPC UA FX over TSN, which utilizes time-sensitive networking (TSN) to further enhance the reliable, low-latency communication required for industrial automation and advanced robotics.
In field trials of quadruped and biped robots traversing outdoor trails, tight timing prevented missteps when the terrain shifted underfoot.
Beyond pure control loops, AI-driven planners predict center-of-mass shifts and adjust joint targets on the fly. These systems blur the line between motion planning and motor control – embedding sensory feedback into every step.
Perception and navigation in complex environments
In a traditional warehouse, multiple 2D LiDAR units and QR-code-reading cameras will often suffice. But in unstructured, human-populated spaces, robots need richer, denser awareness.
Legged and humanoid robots combine 3D LiDAR, time-of-flight (ToF) depth cameras and stereo vision to build volumetric maps in real time.
Simultaneous Localization and Mapping (SLAM) algorithms fuse this data with Inertial Measurement Unit (IMU) readings to maintain accuracy even when visual features are sparse – under hospital curtains or in dim domestic lighting, for instance.
Traditional collision-avoidance treats all obstacles equally. Advanced systems use edge AI to distinguish static furniture from moving humans or pets.
A floor-cleaning robot might pause and reroute when it detects a child’s toy, for example, then resume cleaning once the path clears, minimizing interruption in dynamic settings.
In restaurants or care homes, robots must pick up cups or tools and hand them to people. Grasp planners leverage 3D point clouds and learned object models to identify stable gripping points.
In a simulated kitchen task, a humanoid robot correctly lifted varied vessels 92 percent of the time by combining vision-based detection with force-feedback compensation.
In mixed human-robot workspaces, safety standards demand redundant sensing. Zonal LiDAR scanners enforce protective zones around moving parts, shutting down motion if a person is too close.
While costly, these sensors remain essential in healthcare and hospitality applications.
Modularity and flexibility for rapid deployment
Human-centric robots also need to integrate seamlessly into existing infrastructures, while supporting frequent updates and adapting to evolving tasks.
Autonomous Mobile Robots (AMRs) once relied on a central “brain” to handle all SLAM, vision and control workloads.
Today’s systems push compute to edge modules: Neural Processing Units (NPUs) co-located with cameras for object detection, real-time IMU processing on microcontrollers for stabilization, and multicore hosts for high-level planning.
This distributed approach slashes both overall power consumption and material costs, while delivering optimal performance.
Meanwhile, the Robot Operating System 2 (ROS 2) provides a hardware-agnostic framework for message passing, lifecycle management, and real-time control.
Its support for Data Distribution Service (DDS) transport and actions simplifies coordination among sensor, planning, and actuation nodes, accelerating prototyping and reducing integration risk.
From Wi-Fi in hospitals to LTE or 5G outdoors, robots need flexible networking stacks.
Matter-based protocols may soon unify smart home connectivity, while edge-to-cloud pipelines feed analytics engines for fleet management and predictive maintenance.
Battery management subsystems optimize runtime via dynamic power scaling and health diagnostics, maximizing uptime between charges.
On the actuation side, crossover microcontrollers handle closed-loop motor control and support field-bus interfaces per axle or limb – letting developers scale robot platforms from four to thirty actuators without redesigning core electronics.
Toward mainstream adoption
Legged and humanoid robots are no longer science fiction. Early trials in hospitals are already under way, delivering medication, guiding visitors and sanitizing rooms.
In restaurants, robotic waitstaff are taking orders and clearing tables, while domestic assistants navigate cluttered living rooms to provide errand-running services.
Real-world deployments demand rigorous integration of control, perception, and modular architectures, however. As these systems venture outside perfectly controlled factories, three truths emerge.
First, bipedal and quadrupedal platforms excel where wheels slip, sink or stall, such as soft turf, uneven terrain or cluttered hallways.
Second, multi-modal sensing is increasingly essential. Only by fusing LiDAR, vision and inertial data can robots navigate safely and effectively in human-centric spaces.
Finally, modular compute and standardized middleware are reducing time-to-market and supporting continuous innovation.
In future, the convergence of advanced motion control, richer perception and plug-and-play hardware will usher in a new class of service robots that move seamlessly among people and adapt to the unpredictability of everyday environments.
As these legged platforms become more capable and affordable, we can expect to see them not only in industry, but also in our homes, hospitals, and public spaces, turning the science fiction of years gone by into an everyday reality.