AI
EviRover: Reinforcing Agentic Perception Beyond a Glance
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 29/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 09:34:17 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 03:18:25 UTC.
Briefings
EviRover represents the first perception agent trained to resolve queries through interaction rather than single-glance prediction. It was developed using two data pipelines (EviRover-SFT-5K and EviRover-RL-12K) and a human-verified benchmark (EviLens with 688 instances). The 4B parameter model outperformed its backbone by 30 points on EviLens and achieved 15-point improvements on WebEyes and BrowseComp-VL. All code, models, and data are publicly released.
Key points
- EviRover is the first perception agent explicitly trained to resolve perceptual queries through interaction, utilizing supervised fine-tuning followed by agentic reinforcement learning.
- The 4B parameter EviRover model outperformed its backbone by 30 points on average on the EviLens benchmark, achieving performance comparable to advanced proprietary models.
Context and limitations
Performance gains on EviLens transfer to WebEyes and general multimodal benchmarks, but specific results on other perception tasks remain unverified.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 09:34:17 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeicpekpvo4can3vxc5l4gwkywqrvx7yb54jztgqm2tlvag3t4aef5m