Introduction to WherobotsAI

​

What does WherobotsAI let me do today?

• WherobotsAI Raster Inference - Use Computer Vision to gather insights from raster data at planetary scale.
• WherobotsAI GeoStats - Leverage distributed Machine Learning clustering algorithms to quickly detect hot spots, areas of high activity versus low activity, and local outliers.
• WherobotsAI Map Matching - Leverage distributed Machine Learning systems to match coordinates to maps at scale.

​

Use Cases

​

WherobotsAI Raster Inference

• Land Use Classification: Continuously analyze large-scale overhead imagery to gain a deep understanding of land utilization patterns.
• Infrastructure Detection: Accurately pinpoint and identify infrastructure elements like buildings, solar panels, and offshore turbines within satellite imagery.
• Deforestation Monitoring: Effectively track and measure the rate and extent of deforestation activities.

​

WherobotsAI GeoStats

• Pedestrian Activity Hot spots: Identify areas with high foot traffic within neighborhoods to inform urban planning and business decisions.
• Public Health Outbreak Detection: Pinpoint clusters of disease outbreaks for timely public health interventions.
• Strategic Retail Placement: Analyze competitor locations and customer demographics to optimize store placement for maximum reach and impact.

​

WherobotsAI Map Matching

• Traffic Pattern Analysis: Gain valuable insights into traffic flow and congestion by accurately matching GPS or other location-tracking coordinates to digital maps.
• Optimized Routing: Leverage up-to-date road infrastructure information to determine the most efficient routes between source and destination points.

​

How it works

​

WherobotsAI Raster Inference

​

Computer Vision Model Inference at Planetary Scale

1. Create the WherobotsDB Context.
2. Bring your own computer vision model or use one of our hosted models. WherobotsAI Raster Inference lets you use the following hosted models:
   • Classification - model_id = 'landcover-eurosat-sentinel2'
   • Object detection - model_id = 'marine-satlas-sentinel2'
   • Segmentation - model_id = 'solar-satlas-sentinel2'
   • Text-prompted Object Detection - model_id = 'owlv2'
   • Text-prompted Instance Segmentation - model_id = 'sam2'
3. Load your dataset.
4. Run model inference on your raster dataset.
5. Visualize your results.

Raster inference example

# Example: Running Raster Inference with a Wherobots hosted model for classifying satellite imagery

## Create a WherobotsDB Context
sedona = SedonaContext.create(config)

## Choose a Computer Vision model
model_id = 'landcover-eurosat-sentinel2'

## Choose a dataset
tif_folder_path = 's3a://wherobots-examples/data/eurosat_small'

## Run raster inference
predictions = create_single_label_classification_udfs(model_id, tif_folder_path)

## Visualize results
SedonaUtils.display_image(htmlDF.orderBy(rand()).limit(3))

​

WherobotsAI GeoStats

​

Machine Learning Toolkit for Vector Data Analysis

1. Create the WherobotsDB Context.
2. Load your vector dataset.
3. From Geostats’ distributed Machine Learning Toolkit, choose an algorithm that works best for your research. Geostats offers the following algorithms:
   • DBSCAN
   • Getis-Ord Gi*
   • Local Outlier Factor
4. Cluster your data.
5. Explore your results.

GeoStats Example

# Example, running DBSCAN on data to locate high and low density regions

## Create a WherobotsDB Context
sedona = SedonaContext.create(config)

## Choose your vector dataset
centers = [[1, 1], [-1, -1], [1, -1]]
X, labels_true = make_blobs(
    n_samples=750, centers=centers, cluster_std=0.4, random_state=0
)

## Choose a Machine Learning clustering algorithm and cluster your data
clusters_df = dbscan(df, 0.3, 10, include_outliers=True)

## Explore your results
gdf.plot(
    figsize=(10, 8),
    column="cluster",
    markersize=gdf['isCore'],
    edgecolor='lightgray',
)

​

WherobotsAI Map Matching

​

Offline distributed map matching

1. Create the WherobotsDB Context.
2. Load your location tracking data.
3. Load your map data.
4. Run Map Matching to map your location tracking data.
5. Explore your results.

Map Matching Example

# Example: Running Map Matching on Open Street Map data with GPS coordinates

## Create a WherobotsDB Context
sedona = SedonaContext.create(config)

## Load your location tracking data
dfEdge = matcher.load_osm("data/osm.xml", "[car]")

## Run Map Matching
result_df = matcher.match(df_edge, df_paths, "geometry", "geometry")

## Explore your results
SedonaKepler.create_map()

​

Get Started

1. Create a Wherobots Organization. Different Organization Editions have different capabilities. The following table details which notebooks can be run in each type of Organization. For a detailed comparison of Wherobots Organization Editions, see Wherobots Pricing. To create or join a Wherobots Organization, see Create an Account in the Wherobots Documentation.
   WherobotsAI Raster Inference requires a GPU-Optimized runtime. You can access a GPU runtime by signing up for a paid Wherobots Organization (Professional or Enterprise) and submitting a Compute Request for a GPU-Optimized runtime.

   Notebook	Community Edition	Professional Edition	Enterprise Edition
   Raster Inference - Classification	❌	✅	✅
   Raster Inference - Segmentation	❌	✅	✅
   Raster Inference - Object Detection	❌	✅	✅
   GeoStats - DBSCAN	✅	✅	✅
   GeoStats - LOF	✅	✅	✅
   GeoStats - Gi*	✅	✅	✅
   Map Matching	✅	✅	✅

2. Run a notebook. Sign in to your Wherobots Organization, launch a runtime, and navigate to a notebook. For more information on starting and using notebooks, see Notebook Instance management and Jupyter Instance Management in the Wherobots Documentation.