Research map / Architectures and efficiency / Attention and transformers
[Re] Differentiable Spatial Planning using Transformers
Rohit Ranjan, Himadri Bhakta, Animesh Jha, Parv Maheshwari
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 |
|---|---|---|---|
| increase CNN Encoder layers from 2 to 4 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.88 | Improved |
| increase CNN Encoder layers from 2 to 8 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.90 | Improved |
| reduce d_model from 64 to 32 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.76 | Improved |
| increase d_model from 64 to 128 (ablation) | SPT · M=15 | accuracy: 84.40 -> 85.00 | Improved |
| increase obstacle range from N(0,5) to N(0,10) (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.31 | Hurt |
| increase obstacle range from N(0,5) to N(0,15) (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.67 | Improved |
| reduce attention heads from 8 to 4 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.65 | Improved |
| increase attention heads from 8 to 16 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.24 | Hurt |
| increase Transformer layers from 5 to 8 (ablation) | SPT · M=15 | accuracy: 84.40 -> 84.96 | Improved |
| increase Transformer layers from 5 to 12 (ablation) | SPT · M=15 | accuracy: 84.40 -> 52.33 | Hurt |
| combine CNN layers=8, d_model=128, n_heads=4, and Transformer layers=8 (ablation) | SPT · M=15 | accuracy: 84.40 -> 85.14; 0.87% | Improved |
| use SPT instead of GPPN | Spatial planning · Navigation M=15, M=30, M=50; Manipulation M=18, M=36; Overall | accuracy: 97.10 -> 99.07; 96.17 -> 99.56; 91.97 -> 99.42; 89.06 -> 99.24; 87.23 -> 99.78; 92.31 -> 99.41 | Improved |
Closest papers by meaning
- Differentiable Spatial Planning using Transformers (ICML 2021 · similarity 0.89)
- Symmetry-Aware Transformer Training for Automated Planning (AAAI 2026 · similarity 0.83)
- Planning from Pixels in Environments with Combinatorially Hard Search Spaces (NeurIPS 2021 · similarity 0.81)
- Goal-Space Planning with Subgoal Models (ICML 2025 · similarity 0.80)
- Knot So Simple: A Minimalistic Environment for Spatial Reasoning (NeurIPS 2025 · similarity 0.80)
- FLIP: Flow-Centric Generative Planning as General-Purpose Manipulation World Model (ICLR 2025 · similarity 0.79)
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.
