Research map / Reinforcement learning / Visual and representation RL
PID-Inspired Inductive Biases for Deep Reinforcement Learning in Partially Observable Control Tasks
Ian Char, Jeff Schneider
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 |
|---|---|---|---|
| use PIDE history encoder instead of GRU | SAC · DMSD tracking average | normalized score: 18.25 -> 90.84 | Improved |
| use GPIDE instead of TD3-GRU encoder | SAC · PyBullet average | normalized score: 36.50 -> 63.18 | Improved |
| replace all heads with attention heads (ablation) | GPIDE · PyBullet average | percent difference: -39.44% | Hurt |
| replace all heads with attention heads (ablation) | GPIDE · DMSD average | percent difference: -54.95% | Hurt |
| add one summation head to exponential smoothing heads (ablation) | GPIDE · PyBullet average | percent difference: +11.00% | Improved |
| add one summation head to exponential smoothing heads (ablation) | GPIDE · MSD vs DMSD average | percent difference: -8.33% on MSD, +5.49% on DMSD | Mixed |
| use only exponential smoothing heads (ablation) | GPIDE · DMSD and PyBullet average | percent difference: -11.14% on DMSD, +5.81% on PyBullet | Mixed |
| add VIB to GRU policy | GRU · Navigation Sim/Real | average return: -29.85 -> -39.57 | Hurt |
| add VIB to GRU policy | GRU · MSD Small/Large | average return: -9.82 -> -9.57 | Improved |
| use PIDE instead of GRU | SAC · Tokamak beta_N-Track Sim/Sim | normalized score: 100.00 -> 0.00 | Hurt |
| use GPIDE instead of GRU | SAC · Tokamak beta_N-Track Sim/Real | normalized score: 40.96 -> 55.21 | Improved |
| add positional encoding to transformer encoder | SAC-Transformer · PyBullet | performance | Hurt |
Closest papers by meaning
- Globally Optimal Policy Gradient Algorithms for Reinforcement Learning with PID Control Policies (NeurIPS 2025 · similarity 0.87)
- Bridging State and History Representations: Understanding Self-Predictive RL (ICLR 2024 · similarity 0.86)
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 2024 · similarity 0.85)
- Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (ICLR 2024 · similarity 0.85)
- Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder (AAAI 2025 · similarity 0.85)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (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.
