Abstract
Mutual localization plays a crucial role in multi-robot cooperation. CREPES, a novel system that focuses on six degrees of freedom (DOF) relative pose estimation for multi-robot systems, is proposed in this paper. CREPES has a compact hardware design using active infrared (IR) LEDs, an IR fish-eye camera, an ultra-wideband (UWB) module and an inertial measurement unit (IMU). By leveraging IR light communication, the system solves data association between visual detection and UWB ranging. Ranging measurements from the UWB and directional information from the camera offer relative 3-DOF position estimation. Combining the mutual relative position with neighbors and the gravity constraints provided by IMUs, we can estimate the 6-DOF relative pose from a single frame of sensor measurements. In addition, we design an estimator based on the error-state Kalman filter (ESKF) to enhance system accuracy and robustness. When multiple neighbors are available, a Pose Graph Optimization (PGO) algorithm is applied to further improve system accuracy. We conduct enormous experiments to demonstrate CREPES' accuracy between robot pairs and a team of robots, as well as performance under challenging conditions.
Introduction
For a swarm of robots to collaborate effectively—whether mapping a forest or navigating a warehouse—they must determine their relative positions and orientations, a process known as mutual localization. Traditional methods often struggle in real-world environments: GPS is unavailable indoors, Motion Capture Systems require pre-installed infrastructure, and vision-based SLAM often fails in low-light or featureless areas.
CREPES (Cooperative RElative Pose Estimation System) addresses these challenges by providing a decentralized, infrastructure-free solution. It enables robots to achieve "one-shot" 6-DOF relative pose estimation, allowing a robot to immediately determine a neighbor's full 3D position and orientation from a single frame of data without prior maps.
System Architecture and Methodology
Hardware Design
Each robot in the CREPES system is equipped with a compact sensor suite:
- Active IR LEDs: A disc-shaped array that transmits a unique ID via pulsating light.
- IR Fisheye Camera: A 185° field-of-view camera that detects IR signals while filtering out visible light noise.
- UWB Module: Provides high-precision distance measurements between robot pairs.
- IMU: Captures acceleration and angular velocity to provide critical gravity constraints.
Software Pipeline
The system processes sensor data through several stages to ensure accuracy and robustness. First, ID Extraction uses IR light communication to solve the data association problem, ensuring UWB distance measurements are correctly paired with visual detections. Next, Raw Pose Estimation combines camera direction, UWB distance, and IMU-derived gravity vectors to calculate the 6-DOF pose. To handle dynamic movements, an Error-State Kalman Filter (ESKF) filters sensor noise. Finally, when three or more robots are present, Pose Graph Optimization (PGO) is applied to ensure mathematical consistency across the entire group.
Experimental Results and Evaluation
CREPES was evaluated through extensive real-world testing, including indoor UAV/UGV flights, long-range outdoor tests up to 27.5 meters, and operations in near-total darkness.
- Accuracy: In standard conditions, the system achieved a median position error of 0.13 meters and a median orientation error of 1.16 degrees.
- Robustness: The use of active IR and UWB makes the system virtually immune to ambient lighting conditions, performing reliably in total darkness where traditional cameras fail. It also maintained stability during aggressive maneuvers with roll angles up to 138°.
- Efficiency: The pipeline is optimized for real-time use; ID extraction takes less than 2ms, and pose estimation/filtering requires less than 1ms, supporting high-speed autonomous flight.
Discussion and Key Insights
A primary innovation of CREPES is the dual-purpose use of IR LEDs for both visual markers and communication, eliminating the "identity crisis" common in multi-robot vision. By using the IMU to anchor the system to gravity, the mathematical complexity of finding relative orientation is significantly reduced.
While the system is highly robust, future work will focus on improving performance during prolonged occlusions. The authors suggest that transitioning from single-frame PGO to continuous-time optimization, which incorporates movement history, could further enhance stability. Additionally, efforts to miniaturize the hardware continue, aiming to bring these capabilities to even smaller micro-drone platforms.
Summary updated