.jpg)
The precise localization of autonomous vehicles in environments where GPS is unavailable — such as tunnels, underground parking lots, or urban canyons — remains one of the most complex challenges in mobile robotics. Inertial navigation systems, although robust in the short term, accumulate drift relentlessly without external correction. Traditionally, this correction has been achieved through dedicated infrastructure, expensive external sensors, or complex multi-sensor fusion, which increases costs and hinders mass deployment.
Against this backdrop, a new line of research proposes transforming the vehicle's standard sensors into a virtual speed sensor capable of correcting gyroscope and accelerometer drift. The approach combines a selective state-space model based on Mamba to capture the temporal dynamics of motion, along with evidential deep learning that quantifies prediction uncertainty through a Normal-Inverse-Gamma distribution. This speed estimation with uncertainty is incorporated as a virtual correction into an error-state Kalman filter, enabling inertial navigation to maintain accuracy within 10% of what a dedicated external speed sensor would provide, even during prolonged GPS outages. The system can run in real-time at 40 Hz on edge hardware, making it viable for commercial applications.
Beyond the automotive field, this type of solution illustrates how artificial intelligence can endow critical systems with uncertainty awareness. At Q2BSTUDIO we develop AI for businesses that not only offer predictions but also know when they are uncertain, an essential requirement in sectors such as autonomous logistics, industrial robotics, or infrastructure monitoring. Integrating state-space models like Mamba into production environments also requires a robust and secure cloud architecture; that is why we offer cloud services aws and azure that guarantee scalability and low latency for these deployments.
The ability to build custom applications that incorporate AI agents and advanced analytics — such as Power BI dashboards that visualize estimation uncertainty — enables companies to make informed decisions. Furthermore, cybersecurity plays a fundamental role when these systems connect to critical networks; we conduct audits and pentesting to protect every layer of the software. From conceptualization to production deployment, at Q2BSTUDIO we combine software engineering, artificial intelligence, and good business practices to create solutions that truly make a difference in complex environments.
