Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Efficiently Combining Human Demonstrations and Interventions for Safe Training of Autonomous Systems in Real-Time
Vinicius G. Goecks, Gregory M. Gremillion, Vernon J. Lawhern, John Valasek, Nicholas R. Waytowich
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Combine demonstrations followed by interventions rather than using demonstrations alone | Cycle-of-Learning · AirSim perching, 20 interactions | task completion: 76.25% (± 2.72% std. error) -> 90.25% (± 5.63% std. error) | Improved |
| Combine demonstrations followed by interventions rather than using demonstrations alone | Cycle-of-Learning · AirSim perching, 8 interactions | task completion: 54.00% (± 8.95% std. error) -> 74.75% (± 9.38% std. error) | Improved |
| Use demonstrations followed by interventions instead of demonstrations alone | Cycle-of-Learning · AirSim perching, averaged over presented conditions and datasets | task completion: +12.81% (± 3.61% std. error) | Improved |
| Use demonstrations followed by interventions instead of demonstrations alone | Cycle-of-Learning · AirSim perching, averaged over presented conditions and datasets | human-provided samples: 32.05% (± 3.25% std. error) less human samples | Improved |
| Use demonstrations followed by interventions instead of demonstrations alone | Cycle-of-Learning · AirSim perching, 20 interactions | human-provided samples: 37.79% fewer data samples | Improved |
| Use interventions only instead of demonstrations followed by interventions | Cycle-of-Learning · AirSim perching, 8, 12, and 20 interactions | task completion | Hurt |
| Use PPO with 2 actions and fixed downward throttle instead of the full action space | PPO · AirSim perching, 500 episodes | task completion: 100% | Improved |
| Increase PPO action dimensions from 2 to 3, requiring throttle control | PPO · AirSim perching, 1000 episodes | task completion: 100% -> less than 5% | Hurt |
Closest papers by meaning
- Reconciling Safety and Performance via Dual-Expert Offline Imitation Learning (NeurIPS 2026 · similarity 0.88)
- Learning to Interactively Learn and Assist (AAAI 2020 · similarity 0.88)
- When a Robot is More Capable than a Human: Learning from Constrained Demonstrators (ICLR 2026 · similarity 0.88)
- RLIF: Interactive Imitation Learning as Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Making AI Policies Transparent to Humans through Demonstrations (AAAI 2024 · similarity 0.86)
- CCIL: Continuity-Based Data Augmentation for Corrective Imitation Learning (ICLR 2024 · similarity 0.86)
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.
