# Differentiable Symbolic Planning: A Neural Architecture for Constraint Reasoning with Learned Feasibility

Source: [Differentiable Symbolic Planning: A Neural Architecture for Constraint Reasoning with Learned Feasibility](<https://www.sota2.com/research/paper/69d33e0c7d0172140ec736bb>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-07T05:00:04.607Z

Authors: Venkatakrishna Reddy Oruganti

Published: 2026-02-19T03:38:03.000Z

## Abstract

Neural networks excel at pattern recognition but struggle with constraint reasoning -- determining whether configurations satisfy logical or physical constraints. We introduce Differentiable Symbolic Planning (DSP), a neural architecture that performs discrete symbolic reasoning while remaining fully differentiable. DSP maintains a feasibility channel (phi) that tracks constraint satisfaction evidence at each node, aggregates this into a global feasibility signal (Phi) through learned rule-weighted combination, and uses sparsemax attention to achieve exact-zero discrete rule selection. We integrate DSP into a Universal Cognitive Kernel (UCK) that combines graph attention with iterative constraint propagation. Evaluated on three constraint reasoning benchmarks -- graph reachability, Boolean satisfiability, and planning feasibility -- UCK+DSP achieves 97.4% accuracy on planning under 4x size generalization (vs. 59.7% for ablated baselines), 96.4% on SAT under 2x generalization, and maintains balanced performance on both positive and negative classes where standard neural approaches collapse. Ablation studies reveal that global phi aggregation is critical: removing it causes accuracy to drop from 98% to 64%. The learned phi signal exhibits interpretable semantics, with values of +18 for feasible cases and -13 for infeasible cases emerging without supervision.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Graph Reachability on Reachability 12 to 30 nodes (test)](<https://www.sota2.com/research/sota/graph-reachability-on-reachability-12-to-30-nodes-test.md>): Graph Reachability; Reachability 12 to 30 nodes (test); Overall Accuracy: 82.7.

- [Planning Feasibility on 8x8 to 16x16 Grids (test)](<https://www.sota2.com/research/sota/planning-feasibility-on-8x8-to-16x16-grids-test.md>): Planning Feasibility; Planning 8x8 to 16x16 grids (test); Overall Accuracy: 97.4.

- [Boolean Satisfiability on SAT 10 to 20 variables (test)](<https://www.sota2.com/research/sota/boolean-satisfiability-on-sat-10-to-20-variables-test.md>): Boolean Satisfiability; SAT 10 to 20 variables (test); Overall Accuracy: 96.4.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.02350>)

[PDF](<https://arxiv.org/pdf/2604.02350.pdf>)
