PCS← Back to projects

Research platform · Case study

AI-Assisted TDD for Robotics Learning

A learning environment that brings browser-based programming, automated evaluation, behavioral telemetry, ROS2/Gazebo simulation, and guardrailed AI tutoring into one workflow.

Context
NSF-funded research, University of Nevada, Reno
Role
Graduate Research Assistant
Timeline
August 2022 to present
Users
100+ students
Status
Platform developed, experimental evaluation in progress

The problem

Help that supports learning instead of bypassing it

Robotics learners need timely, contextual help while writing code, interpreting test failures, running simulations, and debugging robot behavior. Traditional lab support does not always scale, while unrestricted generative-AI answers can skip the reasoning students need to practice.

The system

One connected learning loop

01Learner workspaceBrowser and Flutter interface
02Learning APIFlask services and content
03Automated evaluationTests and hidden checks
04Guided recoveryTelemetry and LLM tutoring

ROS2 and Gazebo simulation connect evaluation results to robot behavior. Attempts, failures, and recovery signals provide context for feedback while guardrails keep final solutions hidden.

What I built

From evaluation infrastructure to research evidence

01

Python-based automated evaluation using unit tests, hidden checks, pattern recognition, and real-time failure analysis.

02

Behavioral telemetry for pass rates, attempts, recovery time, edit-run cycles, stuck states, and learning outcomes.

03

ROS2 and Gazebo workflows for simulation-supported robotics exercises.

04

LLM pipelines that analyze code and learning context while guiding students without exposing final solutions.

05

Personalized analytics and recommendation workflows for learners, instructors, and research evaluation.

06

Demonstrations, technical documentation, peer-reviewed papers, and stakeholder communication.

Evaluation

Measured across system and learning outcomes

Measures: accuracy, efficiency, recovery, robustness, and learner experience.

The research uses hypothesis-driven evaluation. Platform development is complete and experimental evaluation is in progress, so preliminary observations are not presented as final conclusions.

Technology

PythonFlaskFlutterDartROS2GazeboOpenAI APIsPrompt engineeringAutomated testingLearning analyticsStatistical analysis

Evidence

Peer-reviewed work

IEEE FIE 2024Unit Testing Framework for Self-Guided Personalized Online Robotics LearningASEE 2024Student-Centered Personalized Robotics Learning Framework