Autonomous Navigation for GPS Denied Environments

This direction is split into other directions: GPS-denied navigation, indoor localization challenge, indoor signal attenuation. Its last version stays readable and citable here.

Version 1 of 1659 papers740 conceptsfrozen 2026-10-04

Most modern navigation relies on an "eye in the sky." Global Positioning Systems (GPS) provide a universal coordinate frame that allows robots to know exactly where they are. However, in the most critical environments—deep tunnels, underwater trenches, dense forests, and indoor disaster zones—this signal vanishes.

Research in this field focuses on how a robot can maintain spatial awareness when it is effectively "blindfolded" from the global grid. The solution is a transition from global reliance to local autonomy, using onboard sensors to build a world from scratch.

The Challenge of Cumulative Drift

Without a global reference, a robot must rely on "dead reckoning"—estimating its current position based on a previously determined position and its movement over time. The fundamental problem is drift. Small errors in sensor readings (like a wheel slip or IMU noise) accumulate. Without a GPS "reset," these errors grow until the robot’s internal map bears no resemblance to reality.

To combat this, researchers like Konstantin S. Yakovlev and Vsevolod Khithov emphasize distributed architectures. In Distributed control and navigation system for quadrotor UAVs in GPS-denied environments, they demonstrate how multiple drones can merge individual, noisy maps into a shared global representation, using visual markers to "anchor" their relative positions and reduce collective uncertainty.

Replacing Satellites with Local Landmarks

If a robot cannot see satellites, it must see the environment. This has led to a heavy reliance on Simultaneous Localization and Mapping (SLAM). The intuition behind SLAM is a feedback loop: the robot uses its sensors to build a map, and then uses that map to figure out where it is.

Different environments require different "eyes":

  • Tunnels and Subterranean Spaces: In featureless, dark environments like metro tunnels, standard SLAM often fails because every section of the tunnel looks like the last. Yusheng Wang 0003 and colleagues developed MetroLoc, which uses a factor graph to prioritize stable geometric features like rail lines and tunnel planes, preventing the algorithm from getting "lost" in the symmetry.
  • Underwater: GPS signals cannot penetrate water. Claudia Soares and Pusheng Ji proposed DIeSEL, which uses pairwise acoustic ranging between vehicles to localize a swarm, treating the network itself as a moving reference frame.
  • Agriculture: Under dense crop canopies, foliage blocks GPS. Yayun Du and Darren Tsang show in Deep-CNN based Robotic Multi-Class Under-Canopy Weed Control in Precision Farming that robots can navigate by identifying specific plant species and rows, using the "messy" biological environment as a structured map.

Learning to Navigate by Mimicry

While geometric SLAM is precise, it is computationally expensive. A parallel direction explores "end-to-end" navigation, where a robot learns to navigate much like a human does: by recognizing patterns rather than calculating coordinates.

Tsuhan Chen demonstrated in Deep Neural Network for Real-Time Autonomous Indoor Navigation that a simple Convolutional Neural Network (ConvNet) can navigate corridors by mimicking an expert pilot. Instead of building a 3D map, the network learns that certain visual patterns (like a vanishing point in a hallway) correlate with "fly forward" commands. Similarly, Paul Drews used LSTMs to predict cost-maps for high-speed off-road driving, allowing vehicles to maintain stability at 27 mph without any global positioning data.

Zero-Infrastructure and Robustness

The ultimate goal of this research is "zero-infrastructure" autonomy—the ability for a robot to be dropped into a completely unknown, signal-denied environment and function immediately.

This requires a shift from fragile, high-precision sensors to robust, adaptive frameworks. For example, Jiun Fatt Chow found that for aerial robots in tunnels, Hector SLAM—which relies purely on LiDAR scan matching—outperforms methods that require wheel odometry, as drones lack ground contact.

By fusing disparate data—LiDAR for structure, IMUs for motion, and cameras for semantics—researchers are creating systems that don't just survive without GPS, but thrive by treating the environment itself as the ultimate navigation signal. As Gaurav S. Sukhatme and Ronald C. Arkin argue, this "science of autonomy" is what will allow robots to move from controlled labs into the complex, unpredictable reality of the physical world.

Papers in this version

Showing 60 of 659 papers.

History & citationVersion 1 · 2026-10-04

Split into other directions · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_autonomous_navigation_for_gps_denied_env_a3019cde,
  title = {Autonomous Navigation for GPS Denied Environments},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/autonomous-navigation-for-gps-denied-environments/rdr_a3019cde1e9f4a36924485923fb2c209},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_autonomous_navigation_for_gps_denied_env_a3019cde_v1,
  title = {Autonomous Navigation for GPS Denied Environments},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/autonomous-navigation-for-gps-denied-environments/rdr_a3019cde1e9f4a36924485923fb2c209/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}