Teaching Vision-Language Agents with Deliberate Practice (DPPO)

Kari Jaaskelainen

21 Nov 2025 — 1 min read

Building capable embodied AI is hard: real-world data is scarce and training is expensive. This paper introduces Deliberate Practice Policy Optimization (DPPO), a coach-like training loop that helps vision-language agents learn more from less.

How it works: the system alternates between learning from examples (to expand skills) and trial-and-error reinforcement learning (to polish them). It spots its own weak spots and focuses practice where it matters most.

Unified view: framed as preference learning for consistent objectives.
Efficiency: maximizes progress with limited, sparse data.
Results: Pelican-VL 1.0 trained with DPPO improves 20.3% over its base and beats open-source 100B-parameter models by 10.6%.
Open source: code and models released for the community.

Bottom line: DPPO is a practical bridge between vision-language models and real-world embodied skills, making versatile agents cheaper and faster to train.

Paper: https://arxiv.org/abs/2511.16602

Paper: https://arxiv.org/abs/2511.16602v1

Register: https://www.AiFeta.com

AI EmbodiedAI Robotics ReinforcementLearning VisionLanguage VLM MachineLearning OpenSource

Meet GRACE: a moral governor for safer, more transparent AI

AI agents are getting powerful—so how do we make sure they do the right thing, not just the effective thing? Meet GRACE, a reason-based moral governor that keeps AI behavior aligned with human norms by separating moral reasoning from goal-driven decision-making. * Moral Module: uses deontic logic and explicit reasons

Why some fine-tuned LLMs miss phishing—and how to fix it

Not all fine-tuned LLMs spot phishing equally. A new study tests Llama 3.1 8B, Gemma 2 9B, and Mistral on high-stakes phishing detection—and uses SHAP and mechanistic interpretability to reveal why models do (or don’t) generalize. * Architecture × data diversity matters: Gemma 2 9B hits state-of-the-art performance (F1

AI that explains itself—by following the science

AI that explains itself—by following the science Black-box AI can be powerful, but it often can’t tell us why it made a decision. Concept Bottleneck Models (CBMs) try to fix that by predicting human-understandable concepts first, then the final answer. The catch: standard CBMs ignore domain-specific cause-and-effect and

Meet GenomAgent: A Team of AI Specialists for Smarter Genomics Q&A

TL;DR: Finding reliable answers in genomics is hard. GenomAgent turns one big AI into a coordinating team of specialists—and beats the current leader by 12% on a key benchmark. Genomic facts live across many databases. Standard chatbots struggle because they can’t flexibly query those sources. GeneGPT added

Read more

Meet GRACE: a moral governor for safer, more transparent AI

Why some fine-tuned LLMs miss phishing—and how to fix it

AI that explains itself—by following the science

Meet GenomAgent: A Team of AI Specialists for Smarter Genomics Q&A