Google DeepMind launches Gemini Robotics 2 to power full-body humanoid control

·1 views

Google DeepMind released Gemini Robotics 2 on July 30, a suite of three AI models designed to control humanoid robots from head to toe. The system separates reasoning from motor execution: Gemini Robotics ER 2 handles high-level planning, while a vision-language-action model translates instructions into physical movements. A third variant, Gemini Robotics On-Device 2, runs the same motor conversion locally on the robot itself. ER 2 is accessible to developers via the Gemini API and Google AI Studio, while motor control models remain limited to select partners. The architecture has been demonstrated on robots from Apptronik and Boston Dynamics, handling multi-step tasks such as walking, grasping, and inter-robot coordination.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why Long-Running AI Agents Fail: The Problem of Context Debt

As AI agents handle multi-step tasks over extended periods, they accumulate what developers call 'context debt' — a buildup of outdated instructions, discarded hypotheses, and redundant data that degrades decision quality. Unlike running out of processing capacity, context debt occurs when temporary execution material is treated as permanent reasoning input, making critical updates harder to distinguish from obsolete ones. Larger context windows only delay the issue without resolving which information is authoritative or which decisions should persist across restarts. Experts recommend separating active reasoning context from durable storage, raw evidence archives, and versioned business artifacts rather than keeping everything in the model prompt. Structured checkpoints that explicitly record accepted findings, rejections, open questions, and permitted next actions are proposed as a practical way to maintain clarity across long-running workflows.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer shares reusable Stripe payment and webhook setup for Next.js 15 SaaS apps

A developer has published a reusable integration pattern for adding Stripe payments to Next.js 15 SaaS applications, covering checkout sessions and subscription handling. The setup centralises Stripe configuration in a single file and uses a Server Action to create hosted Stripe Checkout Sessions, attaching a user ID in metadata for later reference. A key emphasis of the guide is the correct use of webhooks rather than relying on browser-side redirect pages to confirm payment, as redirects can fail or be closed before they fire. The webhook handler verifies Stripe's signature and updates the database only after receiving a confirmed event directly from Stripe's servers. The author presents this as a production-ready pattern reused across multiple projects to avoid common pitfalls in payment integration.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five Key Trends Shaping AI Agents in 2025: Edge Intelligence to Multi-Agent Collaboration

Analysis of Product Hunt and Hacker News data from August 2025 reveals that AI Agent products have moved beyond demo stages into mainstream commercialization, with 8 of 20 daily Product Hunt launches directly related to AI Agents. Three major drivers — privacy concerns, lower latency, and cost efficiency — are pushing AI workloads from cloud servers onto local devices, as seen in products like Open Minis and yapyap. Multi-agent collaboration is emerging as the next frontier, with tools like mpai and Murmell enabling teams of AI agents and humans to work together on shared tasks. Vertical, domain-specific agents in areas such as fitness, advertising, and content creation are gaining traction as more commercially viable alternatives to general-purpose assistants. Simultaneously, a counter-trend is growing on Hacker News, where users are actively seeking AI-free browsers and operating systems, signaling that the industry is entering a phase where users increasingly distinguish between genuinely useful AI and AI added for its own sake.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

JengaStack Launches Managed Odoo 19 Platform with Automated KRA Tax Compliance

Kenyan SaaS startup JengaStack has built a managed ERP platform designed to remove the technical and compliance barriers Kenyan businesses face when adopting Odoo 19. The platform automatically provisions isolated tenant environments pre-configured with Kenyan tax categories, a local chart of accounts, and KRA eTIMS-compliant invoice layouts. A custom M-Pesa integration module enables real-time payment reconciliation directly into Odoo's general ledger, while single sign-on via Authentik OIDC simplifies user access. The underlying infrastructure runs on Microsoft Azure, using Traefik, PgBouncer, and Grafana to handle routing, database scaling, and continuous monitoring. JengaStack is currently rolling out early access to Kenyan businesses, accounting firms, and IT integrators through its website at jengastack.app.

0 comments Read more at DEV Community