AI combines cameras and LiDAR to mark safe footholds and dangerous rock slopes for hikers
A video demonstration posted by Instagram account @thebigflatnow and later reposted by Poonam Soni (@CodeByPoonam) and others shows AI analyzing rocky terrain in real time in hiking/mountaineering scenarios, marking safe footholds for the left and right foot and rejecting edge or unstable areas.
The video is shown from a first-person perspective, with an overlaid interface reading “ANALYSING TERRAIN” and “CONTACT SEARCH,” and it gives percentage assessments of slope, grip and edges (for example, L1 SLOPE CHECK 75%, R2 REJECT EDGE 25%). It uses green outlines to mark recommended footholds (labeled L or R) and red markings for rejected areas. Poonam Soni’s post also describes another application: a high-resolution camera captures cracks and loose rocks, LiDAR builds a 3D terrain map, AI combines the information to judge unstable areas, and repeated scans detect changes. The goal is to identify risks before rocks fall, allowing engineers to inspect dangerous slopes without having to climb them in person, but the system cannot precisely predict when a fall will happen and can only warn in advance.
Similar technologies already exist in geotechnical engineering and robotics. Academic papers use TLS/UAV LiDAR and deep learning, such as PointNet++, to detect discontinuities and changes in rock slopes; industrial systems, such as Geobotica’s GeoLidar and NHAZCA’s IRIS, combine cameras and LiDAR for slope deformation and rockfall monitoring. Robotics research also fuses camera semantic segmentation with LiDAR geometric information to generate traversability maps for hiking-trail navigation. These are all concept demonstrations and visualizations of existing research directions, not the release of a single new product.
Original sources:
Title: AI uses Cameras and Lidars to spot dangerous rock slopes. Publisher: Poonam Soni (@CodeByPoonam) Date of publication: 2026-10-26 URL：https://x.com/CodeByPoonam/status/2114668231797739615 Source nature: firsthand post by the party concerned (including video) Verification status: This reading of the post and video frames was verified on 2026-10-27 Supporting content: video demonstration content, camera + LiDAR + AI description, credit attribution to @thebigflatnow/IG
Title: Slope-Scale Rockfall Susceptibility Modeling as a 3D Computer Vision Problem Publisher: Remote Sensing / MDPI Date of publication: 2023-05-22 URL：https://www.mdpi.com/2072-4292/15/11/2712 Source nature: research or data Verification status: Abstract and key methodological points were read in this review, verified on 2026-10-27 Supporting content: LiDAR point clouds + AI used for rock-slope rockfall susceptibility modeling (historical background, not proof of the specific performance of this demonstration)
ABAB AI Insight
The real-time foothold marking in the video, distinguishing left and right feet and evaluating slope, grip and edges, is a visual demonstration of existing computer vision and 3D perception technologies, similar to the way robotics combines camera semantics and LiDAR geometry to generate traversability maps. The material from @thebigflatnow was reposted across multiple platforms, but it does not disclose a specific model architecture, training data or independently verified accuracy.
In slope monitoring, LiDAR point-cloud change detection and deep learning, such as PointNet++ and PointCNN, have already been used to identify rockfall and deformation from multi-temporal scans, with accuracy reaching about 75% on research datasets. Industrial solutions, such as Geobotica’s automated deformation monitoring system, also integrate cameras and LiDAR to achieve near-real-time 3D surface tracking without requiring personnel to enter dangerous areas. What these systems have in common is reliance on repeated scans to detect changes, rather than single-shot prediction of the exact time of a rockfall.
Compared with hiking-robot navigation research, this demonstration is more focused on personal real-time assistance, whereas academic work, such as traversability maps that fuse semantic segmentation and geometric analysis, is mostly aimed at autonomous robot path planning. The key difference is that the demonstration does not disclose deployment scale or false-positive data, while mature systems are usually complemented by traditional sensors such as GNSS and inclinometers.
Based on the materials reviewed, this visualization lowers the barrier to understanding the concept, but its practical safety value depends on the model’s robustness under real-world variable lighting, vegetation and dynamic conditions. The key observation is whether reproducible open-source implementations or independent third-party test reports appear; for now, it is more of a concept showcase than a hiking safety tool that can be relied on directly.
ABAB News · Cognitive Law
Real-time foothold marking turns 3D perception into a personal visual aid, but it cannot replace experience-based judgment.
Slope warnings rely on repeated scans to detect changes, rather than single-shot prediction of the exact time of a rockfall.