Case Study 1: Adaptive Tutoring with Embodied Agents
Overview
This case study examines AI-assisted tutoring delivered through a humanoid robot. The focus is on measurable learning gains, teacher workload impact, and implementation constraints.
Scenario
Context: Middle-school mathematics (grades 6–8), with periodic small-group instruction.
Intervention: A humanoid robot conducts guided problem-solving sessions for small groups while the teacher focuses on higher-order instruction.
System Components
- Perception: Microphone array + on-device speech-to-text (privacy-first)
- Pedagogy: Socratic questioning with LLM-driven adaptive hints
- Feedback: Immediate formative feedback and logging of common errors
Measured Outcomes
- Learning Gain: Target Cohen's d ≈ 0.3–0.6 (moderate improvement)
- Teacher Workload: 0.5–1 hour/week saved on repetitive questioning
- Engagement: Increased on-task behavior and verbal participation
Evidence Base
Adaptive tutoring yields documented learning gains. Embodied agents increase engagement. However, success depends on pedagogy alignment, interaction fidelity, and session frequency.
Implementation Guidance
- Prototype dialogue in Gazebo/Unity
- Apply FERPA-informed data handling
- Measure ROI across multiple schools
Acceptance Criteria
- Condensed narrative with metrics
- Reproducibility demo included
- Live classroom trial (deferred)
Governance: All student data handling complies with FERPA and Constitution v1.0.0.