Research map / Detection and segmentation / CNN features
End-to-End Egospheric Spatial Memory
Daniel James Lenton, Stephen James, Ronald Clark, Andrew Davison
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 |
|---|---|---|---|
| use ESM instead of the PO2 baseline | Drone Reacher, Ego Acq, Shape · Drone Reacher, Ego Acq, Shape | targets reached: 14.4(6.1) -> 18.3(6.4) | Improved |
| use ESM instead of the PO2 baseline | Drone Reacher, Freeform Acq, Color · Drone Reacher, Freeform Acq, Color | targets reached: 19.1(12.7) -> 16.6(12.9) | Hurt |
| use ESM instead of the PO2 baseline | Manipulator Reacher, Ego Acq, Color · Manipulator Reacher, Ego Acq, Color | targets reached: 13.0(5.9) -> 11.7(5.3) | Hurt |
| use ESM instead of the PO2 baseline | Manipulator Reacher, Freeform Acq, Color · Manipulator Reacher, Freeform Acq, Color | targets reached: 12.5(6.1) -> 11.0(5.8) | Hurt |
| learn encoded features before storing them in ESM rather than directly projecting RGB (ablation) | ESMN · Drone Reacher, Ego Acq, Shape | targets reached: 4.1(3.8) -> 18.3(6.4) | Improved |
| learn encoded features before storing them in ESM rather than directly projecting RGB (ablation) | ESMN · ScanNet test, 60×80 images, 90×180 memory, 16-frame | segmentation accuracy: 54.1(13.8) -> 59.4(12.7) | Improved |
| use an ESM depth map instead of a single depth frame for avoidance | ESMN-RGB policy · Obstacle avoidance | collisions: 10.8(13.0) -> 2.7(5.5) | Improved |
| add untuned spatial auxiliary losses to the LSTM (ablation) | LSTM-Aux · Reacher tasks | targets reached | Hurt |
| transfer an ESMN policy across camera modalities without further training | ESMN · Reacher camera transfer | targets reached: 20.8(7.8) -> 10.4(5.5); 18.3(6.4) -> 0.7(0.9); 16.6(12.9) -> 8.9(6.5); 8.5(11.2) -> 1.0(1.4); 11.7(5.3) -> 7.5(4.9); 4.7 | Mixed |
| train an ESM policy with reinforcement learning rather than imitation learning | ESMN and ESMN-RGB · MR-Seq-Ego-Color | targets reached | Improved |
Closest papers by meaning
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · similarity 0.89)
- Spatially-Aware Transformers for Embodied Agents (ICLR 2024 · similarity 0.88)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.87)
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · similarity 0.86)
- SPA: 3D Spatial-Awareness Enables Effective Embodied Representation (ICLR 2025 · similarity 0.86)
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.
