Academic Research Project

PickPilot: AI-Powered Robotic Picking Platform

A physical 3DoF suction-robot picking platform integrating real-time computer vision, inverse kinematics, and edge-embedded control.

SYS_TELEMETRY: STANDBYCOORD_SYS_3DOF
X: 142.45 mm
Y: 84.12 mm
Z: -22.3 mm
BASE (J1)
45.3°
SHOULDER (J2)
65.0°
ELBOW (J3)
5.0°

Watch the Live Action Demo

See the 3DoF physical arm solve pathing, execute suction picks, and communicate telemetry in real-time.

Core Capabilities

An engineered pipeline coordinating hardware and software layers.

On-Device Computer Vision

Uses YOLO26x optimized and compiled to CoreML, running directly on the team's iPhones to locate and classify objects locally.

Central Brain

Coordinates kinematics, validates safety envelopes, persistent logs, registers camera status, and routes operator demands.

ESP32-S3 Edge Controller

Runs native WebSocket server to digest joint angles and operate physical stepper motors and vacuums with low latency.

Digital Twin Telemetry

Renders physical movements live on a Three.js simulator showing current robotic poses and active detections.

Autonomous Sequence

Fully autonomous loop: triggers detection, solves inverse kinematics, verifies constraints, and executes picking.

Download PickPilot Vision

Transform your iPhone into a high-precision spatial detection sensor. Runs YOLO/CoreML locally and hosts a local REST server.

PickPilot App Store QR Code

Scan to install on iOS device

v1.0.0 • iOS 16.0+

Operational Workflow

How PickPilot links perception to execution in a closed loop.

1. 1. Capture & Object Detection

The iOS Vision service captures camera feed, executes YOLO inference, estimates depth, and returns structured 3D spatial points.

2. 2. Path Planning & Kinematics

Next.js resolves inverse kinematics (IK) for the joint angles, performs safety boundary checks, and logs execution stages.

3. 3. Direct Hardware Execution

Embedded ESP32-S3 parses commands over WebSockets, adjusts steppers, toggles suction, and streams logs back to Mission Control.

CLASS: RED_CAP (98.4%)
Z: 423.5mm
CAMERA_RAW: ACTIVE DETECTION TRACKING

System Architecture

How the modules interact within the local network topology.

iOS Vision App (YOLO/CoreML REST Service)

REST Endpoint / YOLO Client

Next.js Backend (Central Coordinator & database)

Next.js / Trajectory IK Solver

ESP32-S3 Firmware (WebSocket Server)

WebSocket Server / Stepper motor drivers

Mission Control (Next.js operator board)

Next.js operator interface client

Target Use Cases

Flexible robotics platform designed for multiple scopes.

Logistics & Smart Sorting

Simulating high-frequency packaging sorting and conveyor picking setups under industrial configurations.

Robotics System Integration

Testing control schemes, limits, network topology, and sensor options without costly industrial machinery.

Academic & Lab Research

An accessible platform for experimenting with machine learning, path planning, and embedded control models.