In traditional robotics, localization is often treated as an external service. A robot queries a GPS satellite or a pre-installed network of motion-capture cameras to determine its coordinates. However, in environments like collapsed buildings, deep underwater, or extraterrestrial surfaces, these external "anchors" do not exist.
The research direction of infrastructure-free relative localization shifts the burden of spatial awareness from the environment to the agents themselves. Instead of relying on fixed infrastructure, a swarm of robots treats its own members as dynamic reference points. As Gregory Dudek et al. argued in A taxonomy for multi-agent robotics, a collective can overcome individual sensory limitations by using peers as mobile landmarks to facilitate mapping and navigation.
The Swarm as a Dynamic Reference Frame
The most common mechanism for infrastructure-free localization is inter-agent ranging, typically using Ultra-Wideband (UWB) radio. UWB allows robots to measure the distance to their neighbors with centimeter-level precision. However, distance alone does not provide a coordinate system; a single range measurement only places a neighbor somewhere on a sphere.
To turn these distances into a usable map, the authors of Land & Localize: An Infrastructure-free and Scalable Nano-Drones Swarm with UWB-based Localization propose a "land and localize" strategy. A subset of the swarm lands to become stationary anchors, while the remaining mobile units use these temporary reference points to establish a local coordinate frame via multi-dimensional scaling. This allows a swarm to deploy into an unknown area and establish a spatial reference system in seconds.
Resolving Orientation without Global References
A significant challenge in decentralized systems is the lack of a shared "North." Magnetometers are notoriously unreliable indoors or near metal structures. Without a common heading, two robots might agree they are five meters apart but disagree entirely on which direction that five meters represents.
Steven van der Helm et al. demonstrate in On-board range-based relative localization for micro air vehicles in indoor leader-follower flight that a swarm can resolve this ambiguity through ego-motion. By combining inter-agent ranging with shared inertial data (velocity and acceleration), the system makes the relative state observable even without a common heading reference. Similarly, Xu Fang shows in Angle-Displacement Rigidity Theory With Application to Distributed Network Localization that global positions can be recovered using displacement constraints that are invariant to the rotation of local frames, allowing each robot to compute its position within its own orientation.
Mutual Observation and Drift Correction
Even with ranging, individual robot odometry (like wheel encoders or visual flow) inevitably drifts over time. Infrastructure-free systems combat this by using "mutual observation" to perform collaborative corrections.
In EGO-Swarm: A Fully Autonomous and Decentralized Quadrotor Swarm System in Cluttered Environments, Xin Zhou et al. describe a method where drones visually detect their neighbors and compare that perception against the neighbor's broadcasted trajectory. If the visual "sighting" doesn't match the broadcasted data, the drone adjusts its local coordinate frame to compensate for the drift. Junho Choi et al. further refine this in SaWa-ML: Structure-Aware Pose Correction and Weight Adaptation-Based Robust Multi-Robot Localization by using UWB ranges to build a "rigid" geometric structure of the entire group, preventing long-term drift during high-speed maneuvers.
Hierarchical and Passive Coordination
Not all agents in a swarm need to be equally capable. Haoyang Wang et al. propose a hierarchical approach in Aerial Shepherds: Enabling Hierarchical Localization in Heterogeneous MAV Swarms, where a few "Advanced MAVs" act as mobile infrastructure. These "shepherds" monitor and correct the positions of lower-cost, resource-constrained drones, shifting the computational burden away from the bulk of the swarm.
For scenarios requiring stealth or minimal electronic interference, researchers have explored "bearings-only" passive localization. Tingrong Li et al. show in Bearings only passive location of UAV in formation flight based on computer simulation that a drone can correct its position deviation using only the angles (azimuths) of signals from three other drones. This allows for formation maintenance under "electromagnetic silence," where agents minimize active transmissions to avoid detection or interference.
By treating the collective as a self-correcting geometric web, these methods enable robotic swarms to function as autonomous units, independent of the environments they explore.
Key Foundational Work
A taxonomy for multi-agent robotics (1996) — contributed 1 concept elements to this direction.