{"slug": "egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric", "title": "EgoP$^{2}$C: Internalizing Plan-Perceive-Compose for Spatial Reasoning in Egocentric Video", "summary": "A new arXiv paper (2610.22681v1) introduces EgoP$^{2}$C, a framework that trains a single vision-language model to internalize a plan-perceive-compose workflow for spatial reasoning in egocentric video, removing the need for external vision and geometry tools at inference. EgoP$^{2}$C students built on Qwen3-VL-8B and SenseNova-SI-1.3-8B outperform their base models by an average of 9.2 and 7.5 percentage points across three spatial reasoning benchmarks, while cutting inference FLOPs by 65.8-66.3% and achieving 1.29-1.49x speedups over their shared Dual-LoRA teacher. The method first trains specialized Planner and Perception teachers with separate adapters, then fine-tunes mixture-of-experts student experts under their supervision using a segment-level extension of multi-teacher on-policy distillation (MOPD).", "body_md": "arXiv:2610.22681v1 Announce Type: new \nAbstract: Answering complex spatial reasoning questions from egocentric video requires determining what evidence is needed, acquiring targeted observations, and combining them through multiple reasoning steps. Existing vision-language models (VLMs) struggle to perform this process reliably. Agentic systems address this challenge by querying external vision and geometry tools, but training their reasoning leaves these tools unchanged, so inaccurate observations can still lead to incorrect answers. We introduce EgoP$^{2}$C, a framework that internalizes a plan-perceive-compose process within a single VLM. To this end, we first train specialized Planner and Perception teachers using separate adapters (Dual-LoRA), then fine-tune the corresponding experts in a mixture-of-experts (MoE) student under their supervision. We subsequently train the student on trajectories generated by its current policy using a segment-level extension of multi-teacher on-policy distillation (MOPD), assigning different trajectory segments to different supervisors. The Planner teacher supervises which evidence-seeking questions to ask, when to stop gathering evidence, and how to compose the final answer. The Perception teacher supervises the video-grounded answers to those questions. At inference, the student executes this adaptive workflow end-to-end, without external spatial tools. Across three spatial reasoning benchmarks, EgoP$^{2}$C students built from Qwen3-VL-8B and SenseNova-SI-1.3-8B outperform their base models by an average of 9.2 and 7.5 percentage points, respectively. Across both backbones, EgoP$^{2}$C reduces inference FLOPs by 65.8-66.3% and achieves $1.29$-$1.49\\times$ speedups over their shared Dual-LoRA teacher.", "url": "https://wpnews.pro/news/egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric", "canonical_source": "https://arxiv.org/abs/2610.22681", "published_at": "2026-10-19 04:00:00+00:00", "updated_at": "2026-10-19 04:22:02.254789+00:00", "lang": "en", "topics": ["artificial-intelligence", "computer-vision", "ai-research", "large-language-models", "machine-learning"], "entities": ["EgoP$^{2}$C", "Qwen3-VL-8B", "SenseNova-SI-1.3-8B", "arXiv", "Dual-LoRA", "MOPD"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric", "markdown": "https://wpnews.pro/news/egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric.md", "text": "https://wpnews.pro/news/egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric.txt", "jsonld": "https://wpnews.pro/news/egop-2-c-internalizing-plan-perceive-compose-for-spatial-reasoning-in-egocentric.jsonld"}}