Researchers have built a sonar-only navigation system that stops robots from scrambling their own maps when hallways sound alike.
BatSLAM 2.0, described in a new arXiv paper, upgrades the decade-old BatSLAM project, which let robots build maps using biomimetic binaural sonar inspired by echolocating bats. The core problem: sonar echoes from different corridors can look nearly identical, and a wrong match (called a loop closure) can collapse the whole map. The new system adds three pieces: a reworked acoustic front-end, a sequence verifier that checks loop-closure candidates against a run of recent readings instead of a single snapshot, and a pose graph built on a high-performance factor-graph framework. The team tested it on both simulated and real-world recordings and reported robust mapping that resists collapse as map size grows.
This matters because most robots that map their surroundings rely on cameras, using simultaneous localization and mapping (SLAM) systems that need decent lighting and clear sightlines. Sonar-only SLAM skips all that: it can work in smoke, fog, or darkness where cameras struggle, and sonar hardware costs far less than lidar. BatSLAM 2.0's sequence verification borrows a trick familiar from vision-based SLAM, where single-frame matches are also unreliable, and adapts it to audio, a much blunter signal than an image.
It will not replace lidar on a self-driving car, but for a cheap robot feeling its way through a dark basement, borrowing a trick from bats is a smart move.