Research map / Reinforcement learning / RL from feedback
Return-to-Go Is More Than a Number: Q-Guided Alignment for Return-Conditioned Supervised Learning
Yuxiao Yang, Weitong Zhang
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 |
|---|---|---|---|
| add Q-guided RTG-to-behavior alignment (Q-align DT vs QCS) | Decision Transformer · D4RL Gym (sum of 9) | normalized score: 812.3 -> 856.9 | Improved |
| add Q-guided RTG-to-behavior alignment (Q-align DT vs DT) | Decision Transformer · D4RL Gym alignment (mean of 9) | alignment RMSE: 21.39 -> 9.01 | Improved |
| add Q-guided alignment (Q-align DT vs QCS) | Decision Transformer · D4RL AntMaze | normalized score: antmaze-medium-diverse 79.5 -> 76.2; antmaze-medium-play 81.6 -> 85.6 | Mixed |
| add 1D convolution layer (ablation) | Q-align DT · Gym Medium | alignment RMSE: 14.22 -> 9.19 | Improved |
| add 1D convolution layer (ablation) | Q-align DT · Gym Medium-Replay | alignment RMSE: 13.37 -> 11.31 | Improved |
| add 1D convolution layer (ablation) | Q-align DT · Gym Medium-Expert | alignment RMSE: 10.70 -> 6.80 | Improved |
| add 1D convolution layer (ablation) | Q-align DT · Gym Medium-Replay/Medium/Medium-Expert | normalized score: 86.88 -> 86.45; 87.21 -> 87.70; 110.94 -> 111.37 | No clear effect |
| increase RTG perturbation offset Delta RTG from 0 to 10 (ablation) | Q-align DT · halfcheetah-medium | D4RL score | Improved |
| increase RTG perturbation offset Delta RTG from 0 to 10 (ablation) | Q-align DT · halfcheetah-medium | alignment error | Hurt |
| increase number of calibration samples from 5 to 40 (ablation) | Q-align DT · HalfCheetah-vel zero-shot | average episode return: -158.5 -> -142.3 | Improved |
| use Q-align DT in zero-shot transfer instead of QCS | Q-align DT · HalfCheetah-vel zero-shot | average episode return: -1298.7 -> -142.3 | Improved |
Closest papers by meaning
- Adaptive $Q$-Aid for Conditional Supervised Learning in Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Reinformer: Max-Return Sequence Modeling for Offline RL (ICML 2024 · similarity 0.86)
- TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion Alignment (NeurIPS 2026 · similarity 0.85)
- Aligning Diffusion Behaviors with Q-functions for Efficient Continuous Control (NeurIPS 2024 · similarity 0.85)
- Critic-Guided Decision Transformer for Offline Reinforcement Learning (AAAI 2024 · similarity 0.84)
- QPRL : Learning Optimal Policies with Quasi-Potential Functions for Asymmetric Traversal (ICML 2025 · similarity 0.84)
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.
