Research map / Multi-agent systems and games / Human-AI and goal-conditioned
An Interactive Navigation Method with Effect-oriented Affordance
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes ADIN, a modular interactive navigation system that uses a multi-level (object, pose, effect) affordance map to plan long-term paths through obstacles that can be removed by interaction.
Problem
Interactive navigation (InterNav) in cluttered "messy" environments, where obstacles block the path and the agent must decide which obstacles to interact with for efficient long-term navigation; prior RL methods learn only short-term interaction.
Approach
Affordance functions predict object affordances (pushable/pickable), pose affordance (visible), and effect affordance (time cost to remove an obstacle), which are projected into a top-down affordance map; FMM distance on this map is augmented with the time cost so the planner picks the most time-efficient path.
Key results
- ADIN-continuous reaches 59.0% SR on all split, outperforming NIE (50.0%) by 9.0 points and reaching 31.3% SR on hard split vs. NIE's 20.5% SPL-level comparison in Table 1 (SR 37.3 on hard).
- Map-based ADIN variants improve SR by over 40 points over Map+RI (18.5% SR) on the all split.
- Full ADIN with effect affordance reaches 61.2 SR on all split in the component ablation (Table 2).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use effect-oriented affordance planning with ADIN-continuous instead of NIE | Interactive navigation in ProcTHOR · ProcTHOR test (all) | SR: +9.0% SR | Helped |
| Use effect-oriented affordance planning with ADIN-continuous instead of NIE | Interactive navigation in ProcTHOR · ProcTHOR test (hard split) | SR: +8.8% SR | Helped |
| Add pose affordance to object affordance (ablation) | ADIN · ProcTHOR test (all) | SR: 45.9 -> 49.0 | Helped |
| Add pose affordance to object affordance (ablation) | ADIN · ProcTHOR test (hard split) | SR: 36.4 -> 35.9 | Hurt |
| Add time cost to distance-map planning (ablation) | ADIN · ProcTHOR test (all) | SR: 49.0 -> 55.0 | Helped |
| Add time cost to distance-map planning (ablation) | ADIN · ProcTHOR test (hard split) | SR: 35.9 -> 42.0 | Helped |
| Add continuous effect affordance (ablation) | ADIN with time-cost planning · ProcTHOR test (all) | SR: 55.0 -> 59.0 | Helped |
| Add continuous effect affordance (ablation) | ADIN with time-cost planning · ProcTHOR test (hard split) | SR: 42.0 -> 46.1 | Helped |
| Use ground-truth object masks instead of predicted masks (ablation) | ADIN-continuous · ProcTHOR test (all) | SR: 59.0 -> 61.2 | Helped |
| Use ground-truth object masks instead of predicted masks (ablation) | ADIN-continuous · ProcTHOR test (hard split) | SR: 46.1 -> 48.3 | Helped |
Surprising findings
- Object and pose affordances raise interaction accuracy but barely help clearing the path without effect affordance guidance.
Limitations
- Evaluated only in the ProcTHOR simulator with a fixed set of 20 obstacle categories.
- Affordance functions are trained on static data collected with the agent's own interaction policy.
Experimental setup
Datasets: ProcTHOR training set (450k episodes) ProcTHOR static affordance dataset (100k samples) Yolov7 finetuning images (150k)
Benchmarks: ProcTHOR InterNav test split (all and hard)
Metrics: Success Rate Final Distance to Target SPL STS push success rate pick success rate decreased blocked ratio
Baselines: DD-PPO NIE Map+RI ADIN+GP ADIN-discrete
Builds on: Fast Marching Method Semantic map-based ObjectNav (Chaplot et al.) NIE (Zeng et al.) Yolov7
Closest papers by meaning
- Affordances-Oriented Planning Using Foundation Models for Continuous Vision-Language Navigation (AAAI 2025 · similarity 0.89)
- GUI Exploration Lab: Enhancing Screen Navigation in Agents via Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- C-NAV: Towards Self-Evolving Continual Object Navigation in Open World (NeurIPS 2025 · similarity 0.87)
- GAMap: Zero-Shot Object Goal Navigation with Multi-Scale Geometric-Affordance Guidance (NeurIPS 2024 · similarity 0.86)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.86)
- Open-World Reinforcement Learning over Long Short-Term Imagination (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.