End-effector tracking before grasp
This clip isolates the tracking behavior and makes the arm-control stage easier to review.
Demo and evidence
From Detection to Physical Sorting.
This demo shows the robot performing detection, target following, grasping, and categorized placement.
Featured workflows
Bottle and cup sorting runs
These videos capture live detection, arm alignment, dexterous grasping, and final release into the correct category.
Validation
Two object classes
Both bottles and cups appear in the full workflow videos, matching the current scope of the prototype.
Validation
Depth-backed localization
The arm moves against measured 3D target positions rather than only 2D image-space detections.
Validation
Continuous arm alignment
The follow stage keeps the end effector positioned over the target before grasp execution begins.
Outcome
Physical grasp and release
The manipulation sequence closes with a grip, transfer, and categorized placement.
Run summary
What the featured run shows
The main demo now stands on its own in a full-width column so the motion reads cleanly: detect the object, align the arm, grasp with the dexterous hand, then release into the target category.
Primary object
Bottle workflow
The featured clip shows the bottle sequence end to end, from recognition to placement.
Secondary object
Cup workflow
The second clip confirms the same control pipeline can generalize across the two supported classes.
Supporting media
Additional views of the system in motion.
Supporting clips and stills highlight the follow stage, the detection interface, hardware integration, and the physical setup around the robot.
Annotated localization output
This clip shows the visual front end directly, including labels, overlays, and the target used for robot alignment.
Dexterous hand installation
This hardware snapshot shows the integration work behind the grasp sequence seen in the main videos.
Calibration and robot setup
This workspace image ties the demo results back to the calibrated setup that makes physical targeting possible.
1x
Featured robot run
A featured run captures the full detect-follow-grasp-place cycle in a single sequence.
3x
Annotated system frames
Annotated frames separate perception, localization, and manipulation evidence for clear review.
1x
Prototype outcomes
Summarize what is already validated now: bottle/cup recognition, coordinate alignment, dynamic following, and categorized placement.
Support material
The project deck and supporting documents provide additional technical background for the prototype.
Technical source of truth
The repository contains the implementation details behind the demo pipeline.