Research map / Large language models / AI and society
Bidirectional Human-AI Learning in Real-Time Disoriented Balancing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents a real-time system where a human and an AI model mutually assist each other in a visual inverted pendulum balancing task under disorientation, with bidirectional learning via AI fine-tuning from human corrections.
Problem
Human performance degrades under spatial disorientation (as in piloting or spaceflight), and automated assistance can conflict with human preferences and erode trust; the paper asks how humans and AI can adapt to each other in real time.
Approach
A real-time dyadic loop in which an AI assistant gives arrow cues to a disoriented human, and the human's joystick inputs are used to fine-tune the AI (behavior cloning or AIRL), so both partners change behavior; performance is shown with phase portraits before and after each phase.
Key results
- Phase portraits show a sample human decreased oscillation and maintained stability while offset from the direction of balance with AI assistance.
- AI assistants are fine-tuned in about 30 seconds on a consumer laptop for a single training run.
Limitations
- Evaluation is a demonstration with qualitative phase portraits; no quantitative user study is reported.
- The AI assistant pool and cueing thresholds are specific to this task setting.
Experimental setup
Datasets: human subject VIP trial data MARS apparatus human trial data
Metrics: phase portraits of angular velocity vs. angular position crash occurrence
Builds on: soft actor-critic DDPG adversarial inverse reinforcement learning
Closest papers by meaning
- Shared Autonomy with IDA: Interventional Diffusion Assistance (NeurIPS 2024 · similarity 0.82)
- Position: Towards Bidirectional Human-AI Alignment (NeurIPS 2025 · similarity 0.81)
- Making AI Policies Transparent to Humans through Demonstrations (AAAI 2024 · similarity 0.80)
- Maia-2: A Unified Model for Human-AI Alignment in Chess (NeurIPS 2024 · similarity 0.79)
- Discovering Creative Behaviors through DUPLEX: Diverse Universal Features for Policy Exploration (NeurIPS 2024 · similarity 0.79)
- Balancing Humans and Machines: A Study on Integration Scale and Its Impact on Collaborative Performance (AAAI 2024 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.