Hybrid Temporal Straightening: Integrating Neuro-Symbolic Reasoning for Interpretable Latent Planning

TL;DR: What if we combine temporal straightening with rule-based or symbolic reasoning for better interpretability and robustness?

Research Question: Can hybrid models that combine temporal straightening with neuro-symbolic reasoning improve planning interpretability and robustness, especially in sparse or safety-critical settings?

Hypothesis: Hybrid approaches will retain the efficiency and flexibility of neural methods while providing interpretable explanations or symbolic diagnostics for planning failures or risks.

Experiment Plan: - Implement a hybrid model: temporal straightening encoder + symbolic rule learner (e.g., extracting temporal/policy rules from latent trajectories).

• Evaluate on tasks requiring both performance and interpretability (e.g., safety-critical navigation, diagnosis after planning failures).
• Compare interpretability (e.g., rule extraction success), planning quality, and ability to diagnose or correct errors.

References:

• Wang, Y., Bounou, O., Zhou, G., Balestriero, R., Rudner, T. G. J., LeCun, Y., & Ren, M. (2026). Temporal Straightening for Latent Planning.
• Wang, Z., & Li, A. (2025). Temporal Knowledge Graph Reasoning: A Survey of Representation Learning, Rule-based Learning, and Neural-Symbolic Hybrid Methods. International Conference on Data Science in Cyberspace.