Research map / Reinforcement learning / Visual and representation RL
Expressive Temporal Specifications for Reward Monitoring
Omar Adalat, Francesco Belardinelli
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 |
|---|---|---|---|
| replace Boolean monitor with quantitative LTLf[F] monitor | PPO · Acrobot | task completion (%): 4.84% -> 94.77% | Improved |
| replace Boolean monitor with quantitative LTLf[F] monitor | PPO · Pendulum | task completion (%): 45.03% -> 72.11% | Improved |
| replace Boolean monitor with quantitative LTLf[F] monitor | PPO · Bipedal Walker | task completion (%): 7.43% -> 14.43% | Improved |
| replace Boolean monitor with quantitative LTLf[F] monitor | PPO · Cartpole, Mountain Car, Lunar Lander | task completion (%): 30.92% -> 33.34% (Cartpole); 37.04% -> 42.28% (Mountain Car); 43.44% -> 45.61% (Lunar Lander) | Improved |
| replace Boolean monitor with quantitative monitor | Q-learning · Frozen Lake | task completion (%): 62.16% -> 58.96% | Hurt |
| replace Boolean monitor with quantitative monitor | Q-learning · Cliff Walking and Island Navigation | task completion (%): 84.38% -> 84.95% (Cliff Walking); 76.37% -> 76.72% (Island Navigation) | No clear effect |
| replace Boolean monitor with quantitative monitor | Q-learning · Taxi | task completion (%): 46.50% -> 52.76% | Improved |
| replace Boolean monitor with quantitative monitor | Q-learning · Conveyor Belt and Sokoban | task completion (%): 27.1% -> 28.70% (Conveyor Belt); 52.00% -> 53.15% (Sokoban) | No clear effect |
| use quantitative monitor instead of handcrafted reward | Q-learning · Sokoban | task completion (%): 50.23% -> 53.15% | Improved |
| use quantitative monitor instead of handcrafted reward | PPO · Mountain Car | task completion (%): 39.82% -> 42.28% | Improved |
| use quantitative monitor instead of handcrafted reward | PPO and Q-learning · Acrobot, Cartpole, Taxi, Island Navigation, Conveyor Belt | task completion (%): 99.07% -> 94.77% (Acrobot); 71.19% -> 52.76% (Taxi); 96.81% -> 76.72% (Island Navigation) | Hurt |
| use quantitative monitor instead of Boolean monitor | Q-learning · Frozen Lake and Cliff Walking | convergence time (s): 0.03 -> 0.01 (Frozen Lake); 0.04 -> 0.003 (Cliff Walking) | Improved |
Closest papers by meaning
- On the Expressivity of Objective-Specification Formalisms in Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.86)
- Reward Machines for Deep RL in Noisy and Uncertain Environments (NeurIPS 2024 · similarity 0.85)
- Partially Observable Reinforcement Learning with Memory Traces (ICML 2025 · similarity 0.85)
- Reinforcement Learning with Lookahead Information (NeurIPS 2024 · similarity 0.85)
- The Value of Reward Lookahead in Reinforcement Learning (NeurIPS 2024 · 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.
