Research map / Multi-agent systems and games / Human-AI and goal-conditioned
From Gambits to Assurances: Game-Theoretic Integration of Safety and Learning for Interactive Robotics
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.
A research-statement paper by Haimin Hu summarizing his program on game-theoretic safety filters that integrate learning and generative models for interactive robots, without new experiments.
Problem
Ensuring verifiable safety assurances for human-centered robots while minimizing performance loss from safety interventions, under uncertainty about human intent and response.
Approach
Argues for combining game-theoretic safety filters, belief-space planning that closes the safety-learning loop, and generative foundation-model policies blended with game policies via KL penalties, to trade off safety assurance and performance in interactive robotics.
Key results
- Reports a tree-search algorithm for N-robot Stackelberg trajectory games with ~5000x faster computation than brute force and 35% reduction in task completion time versus an order-agnostic dynamic game solver (cited prior work).
- Reports a robust-RL-based neural safety synthesis algorithm that is provably convergent and outperforms prior state of the art on a 36-dimensional quadrupedal locomotion task (cited prior work).
- Reports approximate safety analysis in joint belief-physical space with a 200-dimensional overall state space (cited prior work).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use tree search to compute socially optimal play order and Stackelberg equilibrium strategy | N-robot trajectory games · N-robot trajectory games | computation speed: ∼5000 times faster | Helped |
| Use tree search to compute socially optimal play order and Stackelberg equilibrium strategy | N-robot trajectory games · N-robot trajectory games | task completion time: 35% reduction | Helped |
| Integrate deep reinforcement learning with game theory for neural safety synthesis | Neural safety filter · 36-dimensional quadrupedal locomotion task | performance | Helped |
Limitations
- No new experiments; results are summarized from the author's prior publications.
- Neural safety filters lack safety assurances by design due to black-box nature, as the paper notes.
Experimental setup
Builds on: control barrier function belief-based game-theoretic planning robust reinforcement learning
Closest papers by meaning
- GT-HarmBench: Benchmarking AI Safety Risks Through the Lens of Game Theory (NeurIPS 2026 · similarity 0.86)
- Towards Computational Foreseeability (AAAI 2025 · similarity 0.86)
- Position: AI Safety Must Embrace an Antifragile Perspective (ICML 2025 · similarity 0.85)
- SafeVLA: Towards Safety Alignment of Vision-Language-Action Model via Constrained Learning (NeurIPS 2025 · similarity 0.85)
- IS-Bench: Evaluating Interactive Safety of VLM-Driven Embodied Agents in Daily Household Tasks (AAAI 2026 · similarity 0.85)
- A Physics-Informed Machine Learning Framework for Safe and Optimal Control of Autonomous Systems (ICML 2025 · similarity 0.85)
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.