Research map / Reinforcement learning / Visual and representation RL
Simplified Temporal Consistency Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 temporal consistency to learn policy and value features instead of feeding observations directly | TCRL versus SAC · DMC continuous-control tasks, especially Humanoid and Dog | episodic return | Helped |
| Use temporal consistency representation learning rather than the TD3-like baseline representation | TCRL · Humanoid and Dog DMC tasks | episodic return | Helped |
| Use TCRL instead of the TD3-like baseline | TCRL · Acrobot Swingup and Fish Swim | episodic return | Hurt |
| Use cosine latent consistency loss instead of MSE (ablation) | TCRL-dynamics · Dog Walk | planning performance | Helped |
| Train the value function jointly with latent dynamics using MSE (ablation) | TCRL-dynamics · Dog Run and visual control tasks | performance | Hurt |
| Train the value function jointly with latent dynamics using cosine loss (ablation) | TCRL-dynamics · DMC tasks | performance | Hurt |
| Use the learned dynamics for model-based value expansion (ablation) | TCRL-mve · DMC policy learning tasks | episodic return | Hurt |
| Backpropagate policy updates through the learned dynamics (ablation) | TCRL-bp · Dog Run | episodic return | Hurt |
| Use decision-time MPPI planning with the learned model (ablation) | TCRL-mppi · Acrobot Swingup and Dog Run | episodic return | Helped |
| Replace temporal consistency with observation reconstruction (ablation) | TCRL-rec · DMC state-based tasks, including Dog Walk | episodic return | Hurt |
| Remove the latent-state consistency loss and predict rewards only (ablation) | TCRL-no · DMC state-based tasks, including Dog Walk | episodic return | Hurt |
| Use one-step rather than multi-step prediction during training (ablation) | TCRL-dynamics · Cheetah Run pixel-based control | control performance | Hurt |
Closest papers by meaning
- Efficient Planning with Latent Diffusion (ICLR 2024 · similarity 0.89)
- Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models (NeurIPS 2025 · similarity 0.89)
- Time-Efficient Reinforcement Learning with Stochastic Stateful Policies (ICLR 2024 · similarity 0.88)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.88)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.87)
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 2024 · similarity 0.87)
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.