Synthetic Data Generation for Multimodal Perception call for proposals - Fall 2026
Synthetic Data Generation for Multimodal Perception call for proposals - Fall 2026 supports researchers developing synthetic training data to improve models’ understanding of everyday human activity.
Funder: Amazon Research Awards
Due Dates: November 4, 2026, 11:59 PM Pacific Time (Submission deadline)
Funding Amounts: No more than $80,000 USD cash and $40,000 USD AWS Promotional Credits on average; one-year award period
Summary: Supports synthetic multimodal training-data research that improves activity coverage, cross-modal causal consistency, long coherent episodes, and scenario control.
Key Information: Each principal investigator may submit only one proposal across the ARA Program per call period.
Description
Amazon Research Awards, through Amazon’s Devices and Services organization, supports research into synthetic training data for models that understand everyday human activity. Collecting large, diverse, densely annotated multimodal data from real environments is expensive, slow, and privacy-sensitive. This call seeks approaches that make broader activity coverage, causal consistency across modalities, longer coherent episodes, and control over generated scenarios scale with compute rather than human authoring.
Proposals may address one or more topics: scene-conditioned human motion and object interaction from text; causally coherent audio, speech, and video; hours-long multimodal episodes with persistent world state; and action-conditioned world models producing distinct, plausible outcomes. Theoretical advances, creative ideas, and practical applications are welcome. Rigorous evaluation, including new metrics where standard benchmarks do not exist, and interest in open-source contributions are selection priorities.
