1 of 1

TrashDash

B1: Qimeng Yu, Siying Li, Yilu Huang

18-500 Capstone Design, Spring 2026

Electrical and Computer Engineering Department

Carnegie Mellon University

System Architecture

Product Pitch

TrashDash is a voice-activated, vision-guided mobile trash bin designed for indoor spaces like dorm rooms, where traditional bins are inconvenient to access. With a simple wake word (“Hi TrashDash”), the system activates, detects a user’s hand gesture, and autonomously navigates to the user while avoiding obstacles.

Key Features

  • Voice-controlled activation and stop
  • Real-time hand gesture detection (RGB+Depth camera)
  • Autonomous navigation with obstacle avoidance (ToF sensors)
  • Responsive motor control with multi-directional movement

TrashDash turns a passive object into an interactive, intelligent assistant, improving convenience, hygiene, and accessibility in everyday environments.

Scan to see more information about our project!

https://course.ece.cmu.edu/~ece500/projects/s26-teamb1/

System Description

System Evaluation

Additional Information

Hardware Architecture

Software Architecture

TrashDash is a centralized embedded robotic system built on a Raspberry Pi 5.

The hardware includes

  • OAK-D Pro camera for hand detection
  • Four VL53L1X ToF sensors for obstacle detection via I²C
  • WonderEcho voice module for offline wake-word detection
  • Arduino Uno for ToF sensor connections

The robot uses a four-wheel mecanum chassis with encoder DC motors, driven through a motor driver using GPIO-based PWM signals.

The system is designed to operate under ~25 W power with real-time sensor integration and stable mobile operation.

Metric

Target

Actual

Wake word recognition accuracy within 3m straight-line distance

> 90% accuracy (54/60)

95% (57/60)

Voice command to system response latency

< 1s average latency

0.4s

Hand detection range

Reliable detection from 1-3m

High accuracy 2.5m

70% accuracy at 3m

Scanning rotation finds user

>70% accuracy (7/10)

90% at 1-2.5m

70% at 3m

Battery Life

>= 30 minutes runtime

>1 hour

End-to-end test (no obstacles)

Reaches target zone (<30cm) in >= 85% trials

90%

End-to-end test (with obstacles)

Reaches target zone (<30cm) in >= 75% trials

80%