Blog

Notes on teaching robots new tricks

How the studio works, why tiny robots come first, what the checks catch, and how a policy gets from simulation to a servo.

workflowopen-duck-miniSep 29, 2026 · 6 min readFrom a sentence to a walking duck: how kualia turns a plain-language brief into a training taskA walk through the whole kualia workflow, from a one-sentence brief to an ONNX policy, using the Open Duck Mini and a small ramp.Read the post

Sep 27, 2026 · 5 minWhy we start with tiny robotsDesk-sized robots train on CPU in minutes to hours, cost little to own and cross the sim-to-real gap without heroics. That is why kualia's shelf starts with them.

Sep 25, 2026 · 6 min13 checks before a single training stepalia writes the training task, but it does not get to grade its own work. Every task passes 13 sandbox checks before training can start. Here is what each one catches.

Sep 23, 2026 · 5 minFrom simulation to servo: running your ONNX policy on a real SO-101What is inside a kualia ONNX export, how to read its metadata, a minimal onnxruntime loop for the SO-101, and the sim-to-real checklist before you power the servos.

Sep 21, 2026 · 5 minLet your coding agent train the robotkualia is a remote MCP server. Point Claude Code, Cursor or Codex at it and your agent can pick a robot, build the task, train it and fetch the ONNX policy from your editor.

Research digest

Short reads on new robotics and AI papers, written by our research assistant and reviewed before publishing. 1,060 so far.

Page 55 of 59

Mar 21, 2026Loss of Plasticity as an Optimization Artifact: An Analysis of Fast TRAC Introduction Lifelong reinforcement learning presents a fundamental tension between stability and adaptation. As agents sequentially encounter diverse tasks, they typically exhibit a phenomenon known as loss of plastici Mar 21, 2026 · TRKonuşma Dil Modellerinde Çok Modlu Jailbreak Saldırılarının Optimizasyonu: Güvenlik Değerlendirmelerindeki Kör Nokta Giriş Konuşma Dil Modelleri (Spoken Language Models, SLM), metin tabanlı büyük dil modellerinin (Large Language Models, LLM) yeteneklerini ses işleme ile birleştiren, uçtan uca çalışan yapay zeka sistemleri olarak öne ç Mar 21, 2026 Mar 21, 2026 · TR Mar 21, 2026From Runtime Oracles to Offline Compilers: Reimagining LLM Integration in Lifelong Reinforcement LearningThe integration of large language models into reinforcement learning pipelines has largely followed a predictable script. Researchers treat these models as omniscient advisors, queried continuously during training to pro Mar 21, 2026 · TRKoşullu Dağılımsal Tedavi Etkilerinde Çift Korumalı Tahmin: Teorik Optimalite ile Pratik Kısıtlar Arasında Giriş Nedensel çıkarım literatüründe uzun süredir koşullu ortalama tedavi etkisi (conditional average treatment effect) tahminine odaklanıldı. Ancak tedaviler sadece sonuç değişkeninin ortalamasını değiştirmekle kalmaz; Mar 21, 2026Conditional Distributional Treatment Effects: Statistical Optimality and the Interpretability GapThe dominant paradigm in causal inference focuses on estimating average treatment effects, whether marginal or conditional on covariates. Yet treatments often alter the entire outcome distribution in ways that averages o Mar 21, 2026 · TR Mar 21, 2026Egomotion as Implicit Geometry: Reconsidering 3D Scene Understanding in Multimodal Language ModelsThe pursuit of spatial intelligence in multimodal large language models (MLLMs) has long faced a structural dilemma. On one side, explicit three dimensional representations such as point clouds, bird's eye view maps, and Mar 21, 2026 · TRVisual-ERM: Görsel Eşdeğerlik için Ödül Modellemesi ve Vision-to-Code Görevlerinde Yeni Bir Paradigma Giriş Vision-to-code görevleri, yapay zekanın yapısal görsel anlama yetisinin en zorlu sınavlarından birini temsil etmektedir. Grafiklerin, tabloların veya SVG vektörlerinin executable code veya markdown formatına dönüş Mar 21, 2026Visual Equivalence as the Missing Link in Vision-to-Code Reinforcement Learning Introduction The translation of structured visual inputs into executable code represents one of the most practically consequential frontiers for Large Vision Language Models (LVLMs). Whether converting charts to Python Mar 21, 2026 · TRTELLA ve Sürekli Pekiştirmeli Öğrenmenin Ölçüm Paradoksu Giriş Sürekli pekiştirmeli öğrenme, yapay zeka ajanlarının birden fazla ortam ve görev arasında geçiş yaparken önceki deneyimlerini unutmadan yeni beceriler edinmesini gerektiren temel bir problemdir. Bu alanın en büyük Mar 21, 2026The Measurement Paradox in Continual Reinforcement Learning: Rigor versus Relevance in the TELLA FrameworkThe field of continual reinforcement learning faces a methodological crisis that strikes at the heart of empirical science. We seek to build agents capable of learning throughout their lifetimes, adapting to unpredictabl Mar 21, 2026 · TRGörsel-Dil-Eylem Modellerinde Verimlilik Paradoksu: Çıkarım Verimliliğinden Somutlaştırılmış Verimliliğe Giriş Vision-Language-Action (VLA) modelleri, robotik sistemlerin görsel verileri doğal dil komutlarıyla bütünleştirerek karmaşık manipülasyon görevlerini yerine getirmesini sağlayan yeni bir paradigma olarak öne çıkmak Mar 21, 2026The Embodied Efficiency Gap: Why Smaller Vision-Language-Action Models Can Cost More EnergyIn the relentless pursuit of efficient artificial intelligence, we have grown accustomed to a simple heuristic: fewer parameters and reduced floating point operations translate to lower energy consumption and faster exec Mar 20, 2026 · TR Mar 20, 2026 Mar 20, 2026 · TRYüksek Boyutlu Kaosta Yapısal Düzen: FLUX'un Gizli Renk Alt Uzayı Giriş Metinden görüntüye üretim modellerinde (text-to-image generation) ince taneli kontrol, alanın en temel zorluklarından biri olmaya devam ediyor. Kullanıcılar için üretilen görüntülerin kompozisyonunu, dokusunu veya