Efficient 2D Graph SLAM for Sparse Sensing

Abstract

Simultaneous localization and mapping (SLAM) plays a vital role in mapping unknown spaces and aiding autonomous navigation. Virtually all state-of-the-art solutions today for 2D SLAM are designed for dense and accurate sensors such as laser range-finders (LiDARs). However, these sensors are not suitable for resource-limited nano robots, which become increasingly capable and ubiquitous nowadays, and these robots tend to mount economical and low-power sensors that can only provide sparse and noisy measurements. This introduces a challenging problem called SLAM with sparse sensing. This work addresses the problem by adopting the form of the stateof-the-art graph-based SLAM pipeline with a novel frontend and an improvement for loop closing in the backend, both of which are designed to work with sparse and uncertain range data. Experiments show that the maps constructed by our algorithm have superior quality compared to prior works on sparse sensing. Furthermore, our method is capable of running in real-time on a modern PC with an average processing time of 1/100th the input interval time.

Tiny-PULP-Dronets: Squeezing Neural Networks for Faster and Lighter Inference on Multi-Tasking Autonomous Nano-Drones

“…State-of-the-Art (SoA) perception algorithms, such as simultaneous localization and mapping (SLAM) [13], can not run onboard nano-drones due to their steep performance requirements. Even when run off-board, they suffer from substantial performance degradation as nano-drones employ low-quality sensors [14], [15].…”

Towards a Multi-Pixel Time-of-Flight Indoor Navigation System for Nano-Drone Applications

“…Notably, [37], [40] achieve their objectives without relying on external infrastructure. However, within the nano-UAV domain, even fewer works tackle the mapping challenge [6], [19], [41], and they offload the computation to an external base station. Furthermore, the existing works performing mapping with nano-UAVs are not able to reach the same level of accuracy as standard-size UAVs within the literature.…”

“…Hence, upon revisiting a location (i.e., loop closure), the pose error is higher than at the initial visit. To mitigate this issue, the robot also acquires environmental observations (i.e., depth measurements) during the flight [19]. By comparing the observations associated with two different poses, an accurate rigid body transformation can be derived between the two, using an approach called scan-matching [19].…”

“…In most common applications, the observations consist of depth measurements, typically provided by LiDARs or stereo cameras [21], [22]. Although SLAM paired with LiDARs is widely used in applications with standard-size UAVs, these solutions require large amounts of computational resources and memory, which are not available on nano-UAVs [6], [19]. Furthermore, even the most compact LiDARs used with standard-size UAVs are about one order of magnitude heavier 1 than the maximum payload of nano-UAVs [23].…”

“…The recent release of lightweight (i.e., 42 mg), lowresolution, and energy-efficient depth sensors based on Time of Flight (ToF) technology has changed the status quo in the feasibility of SLAM for nano-UAVs [24]. With the aid of such sensors, recent works demonstrated SLAM on nano-UAVs, but only under the assumption that the complex SLAM could be offloaded to an external base station [19]. This approach reduces the flight time due to the significant power consumption introduced by the radio communication with the base station [7].…”