Datatypes
Datatypes
Skills and Agents need their data in a consistent shape to validate, serialize, and exchange it across the Telekinesis ecosystem. Importing telekinesis.datatypes gives you one common set of types - primitives, geometry, robotics, vision, and models - so one Skill's output slots straight into the next Skill's input, no conversion glue required.
Primitives
Use these when a Skill's input or output is a bare scalar, timestamp, or untyped array with no geometric or domain structure.
Bool
Boolean scalar.
View →Int
Integer scalar.
View →Float
Floating-point scalar.
View →String
String scalar.
View →Timestamp
Clock-epoch sec/nanosec pair, no timezone semantics.
View →DateTime
Timezone-aware wall-clock datetime, normalized to UTC.
View →Array
Arbitrary-shape, arbitrary-dtype NumPy array.
View →Geometry
Use these for the 2D/3D shapes, vectors, matrices, and boxes that describe positions, extents, and detections within vision or robotics data.
Vector2D
2D vector.
View →Vector3D
3D vector.
View →Vector4D
4D vector.
View →Vectors2D
Batch of 2D vectors.
View →Vectors3D
Batch of 3D vectors.
View →Vectors4D
Batch of 4D vectors.
View →Mat2x2
2×2 matrix.
View →Mat3x3
3×3 matrix.
View →Mat4x4
4×4 matrix.
View →Box2D
Single axis-aligned 2D box.
View →Box3D
Single axis-aligned 3D box.
View →OrientedBox2D
Single rotated 2D box.
View →OrientedBox3D
Single rotated 3D box.
View →Boxes2D
Batch of axis-aligned 2D boxes.
View →Boxes3D
Batch of axis-aligned 3D boxes.
View →OrientedBoxes2D
Batch of rotated 2D boxes.
View →OrientedBoxes3D
Batch of rotated 3D boxes.
View →Point2D
Single 2D point.
View →Point3D
Single 3D point.
View →Points2D
Batch of N 2D points.
View →Points3D
Batch of N 3D points.
View →Circle
Single 2D circle.
View →Circles
Batch of 2D circles.
View →Contour
Single 2D contour (polyline).
View →Contours
Batch of 2D contours.
View →Eigenvalues
Eigenvalues of a covariance/matrix.
View →Eigenvectors
Eigenvectors of a covariance/matrix.
View →Robotics
Use these to carry positions, poses, transforms, and motion state between Skills that operate in Cartesian or joint space.
Position2D
2D position (x, y).
View →Position3D
3D position (x, y, z).
View →Transform2D
2D homogeneous transform.
View →Transform3D
3D homogeneous transform.
View →Transforms3D
Batch of 3D homogeneous transforms.
View →Quaternion
Unit quaternion rotation.
View →Pose2D
Single 2D pose.
View →Pose3D
Single 3D pose.
View →Poses2D
Batch of 2D poses.
View →Poses3D
Batch of 3D poses.
View →Twist3D
3D linear + angular velocity.
View →Wrench3D
3D force + torque.
View →Covariance6x6
6×6 pose covariance.
View →OccupancyGrid
2D occupancy grid map.
View →Vision
Use these to represent what a camera or depth sensor perceives - images, point clouds, meshes - and the detections/annotations produced from them.
Image
Image with optional on-wire compression.
View →ImageBatch
Batch of images.
View →SegmentationImage
Per-pixel segmentation mask.
View →DepthImage
Per-pixel depth map.
View →CameraCalibration
Camera intrinsics, resolution, and distortion model.
View →PointCloud
Single 3D point cloud (points + optional colors/normals).
View →PointCloudBatch
Batch of point clouds.
View →Mesh3D
3D triangle mesh.
View →Mesh3DBatch
Batch of independently-sized 3D triangle meshes.
View →Color
RGBA color.
View →COCOObjectDetectionResult
Single COCO-style detection with a confidence score.
View →COCOObjectDetectionResults
Batch of COCO-style detections.
View →COCOObjectDetectionAnnotation
Single COCO-style ground-truth annotation.
View →COCOObjectDetectionAnnotations
Batch of COCO-style ground-truth annotations.
View →Category
Single category definition.
View →Categories
Batch of category definitions.
View →Models
Use these to describe an ML model's input/output tensor signature so every Skill that loads or serves it agrees on shape and dtype.
ModelTensorDefinition
One model input/output tensor signature.
View →ModelDefinitions
Batch of ML model metadata records.
View →Where to Go Next?
Continue to the next tutorial.
Inter-Device Communication
Pub/sub and client/server messaging across devices with BabyROS.
Next tutorial →