Research map / Reinforcement learning / Visual and representation RL
Efficient Policy Adaptation with Contrastive Prompt Ensemble for Embodied Agents
Wonje Choi, Woo Kyung Kim, SeungHyun Kim, Honguk Woo
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 ConPE instead of EmbCLIP | EmbCLIP · AI2THOR ObjectNav unseen | success rate: 59.0 -> 79.7 | Improved |
| use ConPE instead of EmbCLIP | EmbCLIP · Egocentric-Metaworld Reach-Wall unseen | success rate: 49.3 -> 67.3 | Improved |
| use ConPE instead of EmbCLIP | EmbCLIP · CARLA Map 1 unseen | sum of rewards: 1415.1 -> 1933.4 | Improved |
| use ConPE instead of ATC on PointNav seen | ATC · AI2THOR PointNav seen | success rate: 89.1 -> 89.7 | Improved |
| add ConPE prompt ensemble to pretrained policy | Pretrained policy · AI2THOR ObjectNav unseen | success rate: 65.8 -> 72.8 | Improved |
| add ConPE prompt ensemble to pretrained policy | Pretrained policy · Egocentric-Metaworld Button-Press target | success rate: 16.8 -> 73.7 | Improved |
| increase number of prompts from 2 to 10 (ablation) | ConPE · AI2THOR ObjectNav unseen | success rate: 43.0 -> 79.7 | Improved |
| increase number of prompts from 10 to 16 (ablation) | ConPE · AI2THOR ObjectNav unseen | success rate: 79.7 -> 77.1 | No clear effect |
| increase number of prompts from 10 to 18 (ablation) | ConPE · AI2THOR ObjectNav unseen | success rate: 79.7 -> 79.0 | No clear effect |
| replace guided attention with uniform average of prompted embeddings (ENS-UNI-AVG) (ablation) | ConPE · AI2THOR ObjectNav unseen | success rate: 65.5 -> 79.7 | Improved |
| replace guided attention with prompt-level uniform average (COM-UNI-AVG) (ablation) | ConPE · AI2THOR ObjectNav unseen | success rate: 43.1 -> 79.7 | Improved |
| remove policy prompt p_pol from adaptation of pretrained policy (ablation) | ConPE · AI2THOR PointNav target | success rate: 54.2 -> 84.4 | Improved |
Closest papers by meaning
- Selective Visual Representations Improve Convergence and Generalization for Embodied AI (ICLR 2024 · similarity 0.88)
- GenRL: Multimodal-foundation world models for generalization in embodied agents (NeurIPS 2024 · similarity 0.87)
- Act to See, See to Act: Diffusion-Driven Perception-Action Interplay for Adaptive Policies (NeurIPS 2025 · similarity 0.87)
- Robot-R1: Reinforcement Learning for Enhanced Embodied Reasoning in Robotics (NeurIPS 2025 · similarity 0.87)
- HASARD: A Benchmark for Vision-Based Safe Reinforcement Learning in Embodied Agents (ICLR 2025 · similarity 0.87)
- Pre-trained Text-to-Image Diffusion Models Are Versatile Representation Learners for Control (NeurIPS 2024 · similarity 0.87)
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.
