Research map / Architectures and efficiency / RNNs and normalization
Scaling up and Stabilizing Differentiable Planning with Implicit Differentiation
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 |
|---|---|---|---|
| Replace algorithmic differentiation with implicit differentiation in ConvGPPN | ConvGPPN · visual navigation | success rate (%): 81.76 -> 98.91 | Helped |
| Replace algorithmic differentiation with implicit differentiation in ConvGPPN | ConvGPPN · workspace manipulation | success rate (%): 70.23 -> 92.60 | Helped |
| Replace algorithmic differentiation with implicit differentiation in SymVIN | SymVIN · 18×18 configuration-space manipulation | success rate (%): 55.15 -> 99.63 | Helped |
| Replace algorithmic differentiation with implicit differentiation in SymVIN | SymVIN · visual navigation | success rate (%): 96.04 -> 86.41 | Hurt |
| Replace algorithmic differentiation with implicit differentiation in VIN | VIN · 36×36 configuration-space manipulation | success rate (%): 74.75 -> 56.27 | Hurt |
| Use implicit differentiation instead of algorithmic differentiation | differentiable planners · 2D navigation, 49×49 maps | backward-pass runtime | Helped |
| Use implicit differentiation instead of algorithmic differentiation | differentiable planners · 2D navigation, larger maps and iteration counts | scalability | Helped |
| Increase forward iterations to solve the equilibrium adequately (ablation) | implicit differentiable planners · long-horizon planning | success rate | Helped |
| Use Anderson solver instead of forward iteration for the backward pass (ablation) | implicit differentiable planners · planning tasks | success rate | Helped |
| Add Jacobian regularization (ablation) | ID-VIN · 2D navigation | success rate | Hurt |
| Replace the planner update with a self-attention layer in ID-SPT (ablation) | SPT · planning tasks | performance | Hurt |
| Train ConvGPPN with 30 layers on 49×49 maps (ablation) | ConvGPPN · 2D navigation, 49×49 maps | success rate | Hurt |
Closest papers by meaning
- Scaling Value Iteration Networks to 5000 Layers for Extreme Long-Term Planning (ICML 2025 · similarity 0.89)
- Towards Real-World Navigation With Deep Differentiable Planners (CVPR 2022 · similarity 0.87)
- DNCs Require More Planning Steps (ICML 2024 · similarity 0.84)
- Versatile Navigation Under Partial Observability via Value-guided Diffusion Policy (CVPR 2024 · similarity 0.84)
- Tree-Guided Diffusion Planner (NeurIPS 2025 · similarity 0.83)
- Sparse Planning in Visual World Models via Cost Gradients (NeurIPS 2026 · similarity 0.83)
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.