Scalar-Measurement Attitude Estimation on $\mathbf{SO}(3)$ with Bias Compensation
A new method uses just 2-3 scalar measurements instead of full 3D vectors for precise attitude estimation.
A research team including Alessandro Melis, Tarek Bouazza, and Soulaimane Berkane has introduced a groundbreaking method for estimating the 3D orientation (attitude) of robots, drones, and autonomous systems. Their paper, 'Scalar-Measurement Attitude Estimation on SO(3) with Bias Compensation,' challenges the conventional reliance on full vector measurements from sensors like accelerometers and magnetometers. Instead, they propose a framework of nonlinear deterministic observers that operate directly on the SO(3) rotation group, requiring only scalar measurements—individual components of sensor readings or independent constraints. A key breakthrough is the formal proof that, under suitable motion (excitation), just two such scalar measurements are sufficient for observability, and three suffice in a static case, fundamentally changing the requirements for reliable orientation estimation.
The technical framework incorporates real-time gyroscope bias compensation and guarantees uniform local exponential stability. Crucially, it demonstrates robustness to 'partial sensing,' where estimation remains accurate even when only a subset of vector components is available, a common scenario in sensor failures or constrained designs. Experimental validation on the public BROAD dataset showed consistent performance across progressively reduced measurement configurations, with errors staying low even under severe information loss. This positions scalar-measurement-based observers as a practical, reliable, and potentially more resilient alternative to traditional vector-based approaches, with significant implications for cost reduction, sensor miniaturization, and system robustness in aerospace, robotics, and wearable technology.
- Proves only 2 scalar measurements under motion (3 if static) are needed for full 3D attitude estimation, a fundamental advance in observability.
- Validated on BROAD dataset, maintaining small errors even with severe measurement loss, demonstrating robustness to partial sensor failure.
- Incorporates gyroscope bias compensation within nonlinear observers on SO(3), ensuring stability and practical reliability for real systems.
Why It Matters
Enables more fault-tolerant, cost-effective, and miniaturized navigation systems for drones, robots, and AR/VR by drastically reducing sensor data requirements.