Search papers, labs, and topics across Lattice.
To overcome the limitations of fixed-context, single-pass inference in complex video understanding, the authors developed VLX-VR, an agentic multimodal model that actively queries context via an iterative Think鈥揗emory鈥揙bservation loop. This formulation allows the system to adaptively retrieve temporal evidence, maintain working state, and decide when to terminate rather than passively ingesting truncated video streams. Trained with reinforcement learning over multimodal trajectories, VLX-VR achieves a state-of-the-art 78.79% accuracy on MINERVA with remarkably stable cross-duration performance (variance of 2.97 pp虏) and 96.20% reasoning trace consistency.
Instead of forcing entire videos into a single, lossy feed-forward context, training an agent to actively read and write temporal memory yields 78.8% accuracy on MINERVA with virtually zero performance drop on long-duration clips.
Real-world video understanding requires integrating visual, audio, textual, and temporal evidence distributed across a video. Yet many pipelines use a fixed video context and single-pass inference, limiting adaptive evidence acquisition when observations are incomplete, ambiguous, or conflicting. We present VLX-VR, an agentic-aware video reasoning model trained within a video reasoning framework defined by a Think--Memory--Observation loop. At each step, VLX-VR determines the needed evidence, invokes read_memory or write_memory, incorporates the returned Observation, and decides whether to continue or produce the task output. We train VLX-VR with multimodal data, including videos and agent trajectories, using reinforcement learning to learn evidence acquisition, memory use, and termination. On MINERVA, VLX-VR achieves state-of-the-art performance among the models included in our comparison, with 78.79% accuracy. Under the original three duration groups, its accuracies are 76.70%, 78.73%, and 80.92%, with a cross-duration accuracy variance of 2.97~$\mathrm{pp}^2$. On correctly answered samples, 96.20% of VLX-VR's reasoning traces are consistent with the MINERVA reference reasoning traces and the evidence described by them, while approximately 75.80% of all evaluated samples satisfy both answer correctness and this evidence-grounded trace criterion. These results show strong performance and broadly stable behavior across durations, while counting, state changes, causal reasoning, and spatial perception remain challenging.
