返回首页
最新
TLDR: I am using AI&more to make robotic teleoperation faster and sustainable over long periods, enabling large real robotic data collection for robotic foundational models.<p>We are probably 5-6 orders of magnitude short of the real robotic data we will need to train a foundational model for robotics, so how do we get that? I believe simulation or video can be a complement, but there is no substitution for a ton of real robotic data.<p>I’ve been exploring approaches to scale robotic teleoperation, traditionally relegated to slow high-value use cases (nuclear decommissioning, healthcare). Here’s a short video from a raw testing session (requires a lot of explanation!):<p><a href="https://youtu.be/QYJNJj8m8Hg" rel="nofollow">https://youtu.be/QYJNJj8m8Hg</a><p>What is happening here?<p>First of all, this is true robotic teleoperation (often people confuse controlling a robot in line-of-sight with teleoperation): I am controlling a robotic arm via a VR teleoperation setup without wearing it, to improve ergonomics, but watching at camera feeds. Over wifi, with a simulated 300ms latency + 10ms jitter (international round trip latency, say UK to Australia).<p>On the right a pure teleoperation run is shown. Disregard the weird “dragging” movements, they are a drag-and-drop implementation I built to allow the operator to reposition the human arm in a more favorable position without moving the robotic arm. Some of the core issues with affordable remote teleoperation are reduced spatial 3D awareness, human-robot embodiment gap, and poor force-tactile feedback. Combined with network latency and limited robotic hardware dexterity they result in slow and mentally draining operations. Often teleoperators employ a “wait and see” strategy similar to the video, to reduce the effects of latency and reduced 3D awareness. It’s impractical to teleoperate a robot for hour-long sessions.<p>On the left an AI helps the operator twice to sustain long sessions at a higher pace. There is an "action AI" executing individual actions such as picking (the “action AI” right now is a mixture of VLAs [Vision Language Action models], computer vision, motion planning, dynamic motion primitives; in the future it will be only VLAs) and a "human-in-the-loop AI", which is dynamically arbitrating when to give control to the teleoperator or to the action AI. The final movement is the fusion of the AI and the operator movement, with some dynamic weighting based on environmental and contextual factors. In this way the operator is always in control and can handle all the edge cases that the AI is not able to, while the AI does the lion share of the work in subtasks where enough data is already available.<p>Currently it can speed up experienced teleoperators by 100-150% and much more for inexperienced teleoperators. The reduction in mental workload is noticeable from the first few sessions. An important challenge is speeding up further vs a human over long sessions. Technically, besides AI, it’s about improving robotic hardware, 3D telepresence, network optimisation, teleoperation design and ergonomics.<p>I see this effort as part of a larger vision to improve teleoperation infra, scale up robotic data collection and deploy general purpose robots everywhere.<p>About me, I am currently head of AI in Createc, a UK applied robotic R&D lab, in which I built hybrid AI systems. Also 2x startup founder (last one was an AI-robotics exit).<p>I posted this to gather feedback early. I am keen to connect if you find this exciting or useful! I am also open to early stage partnerships.
Show HN: Enfiy Code – Universal AI coding assistant with multi-provider support<p>Hi HN! I built Enfiy Code, a command-line AI coding assistant that works with multiple AI providers (Anthropic Claude, OpenAI GPT, Google Gemini, Ollama for local models, etc.) from a single interface.<p>Key features:
• Switch between AI providers seamlessly - ompare responses from different models
• Works with large codebases using extended context support
• Supports both cloud AI (powerful) and local AI (private) via Ollama
• Integrates external tools through MCP (Model Context Protocol)
• Generate apps from PDFs/sketches using multimodal AI
• Auto-handles complex tasks like PR reviews and git operations<p>The CLI is built with TypeScript/Node.js and is fully open source (Apache 2.0). You can try it without installing: `npx @enfiy/enfiy-code`<p>What makes it different from other AI coding tools is the provider flexibility - you're not locked into one AI service, and you can run everything locally if privacy is a concern.<p>Would love feedback from the HN community, especially on the multi-provider approach and MCP integrations!<p>GitHub: <a href="https://github.com/enfiy/enfiy-code">https://github.com/enfiy/enfiy-code</a>
I just created a quiz that reveals how corporate legal language can trap communities into losing control of their environmental data.<p>It is based on real terms and conditions typical for air quality monitoring manufacturers.<p>Try it and see if you can spot the red flags:
<a href="https://www.airgradient.com/aq-data-ownership-quiz/" rel="nofollow">https://www.airgradient.com/aq-data-ownership-quiz/</a><p>The aim of the game is to create awareness among data ownership rights, e.g.:<p>- How "joint ownership" isn't really sharing
- Why "free" services often delete your data without warning
- How subscription models hold your community's air quality data hostage
- The difference between data access and data ownership<p>The game was inspired by the incredible work of the EPIC Air Quality Fund, which is fighting to expand access to air quality data. Their research revealed that nearly 40% of countries lack open air quality data—often because restrictive corporate terms prevent communities from truly owning their environmental information.<p>At AirGradient, we believe open source hardware is the ultimate solution. When communities control both the hardware AND the data, they achieve true environmental justice.
In large-scale ad delivery, cross-border teams often face "platform barrier" challenges: Materials used for Facebook ads need manual resizing to work on Instagram; TikTok targeting parameters can’t sync to Google Ads; and over 60% of traffic comes from a single channel with no easy way to diversify. These issues cut delivery efficiency by 40% and may even expose teams to traffic fluctuation risks due to channel dependency. Yajuzhen Cloud Phone enables synchronization of materials, audiences, and data across 6 major ad platforms (Facebook, TikTok, Google, etc.) through a "cross-platform collaboration hub + localized delivery engine." This boosts multi-channel traffic share from 40% to 85% and increases delivery ROI by 50%.
Localized Material Generation: Core materials auto-generate multi-language variants. For Germany, English keywords pair with German annotations; for Southeast Asia, Malay voiceovers and local backdrops (e.g., Kuala Lumpur Twin Towers) are added. A 3C brand increased localized material CTR by 28% with this feature.
II. Cross-Platform Audience Strategy Sync: Cut Redundant Testing Costs:
Cross-Platform Tag Mapping: Core traits like "females 25-35 who bought yoga mats" auto-convert to platform-specific tags: "Yoga Equipment" on Facebook, "yoga mat review" on Google, and "yoga tutorial views" on TikTok. A sports brand cut new platform testing from 7 days to 2.
Audience Segment Sync: When Facebook’s "high-conversion audience" gains 100,000 users, the system syncs identical profiles to Google Ads and TikTok, adjusting sizes per platform rules (e.g., Google retains 80% overlap; TikTok expands 20% lookalikes). A beauty brand increased multi-platform audience overlap from 15% to 42%.
III. Intelligent Delivery Rhythm Coordination: Balance Channel Traffic Share
Over-reliance on one channel (e.g., >60% traffic) poses risks: Algorithm changes can crash traffic, and weakened bargaining power raises acquisition costs. Yajuzhen dynamically adjusts delivery across platforms via an "intelligent traffic allocation hub":
Threshold Alerts & Auto-Diversion: When a platform exceeds a 50% traffic share, the system cuts its budget by 20% while boosting bids (e.g., +10% for Google Ads). A cross-border retailer reduced single-channel dependence from 65% to 30%.
Time-Segmented Delivery: Strategies align with platform peak times: Facebook ramps up 20:00-22:00 (Europe/America evenings), TikTok focuses 12:00-14:00 (lunch breaks), and Google covers 0:00-6:00 (late-night searches). This boosts daily traffic utilization by 35%.
IV. Cross-Platform Data Asset Integration: Break Channel Data Silos
Ad data is siloed in platform backends, hindering unified analysis: Facebook’s "conversion paths" differ from TikTok’s "engagement funnels"; Google’s "search term reports" can’t link to Instagram’s "material performance." Yajuzhen integrates data via a "global data center":
Multi-platform material sync cut launch cycles from 5 days to 1.5, enabling 6-platform simultaneous releases;
Audience tag mapping cut Google/TikTok testing costs by 40%, adding 2 core channels;
The traffic allocation system reduced Facebook to 35% share, with TikTok+Google hitting 45%;
The data center identified Instagram video materials as highest-converting, lifting overall ROI by 62%.
The brand manager stated: "Yajuzhen doesn’t just solve ‘multi-platform hassle’—it builds ‘traffic ecosystem resilience.’ When one channel fluctuates, others fill the gap. This risk resistance is critical in cross-border markets."
Yajuzhen Cloud Phone’s core value lies in upgrading "decentralized delivery" to "collaborative operations": breaking platform barriers to let materials, audiences, and data flow freely across channels. This creates a cycle of traffic complementarity, risk diversification, and efficiency gains. For brands pursuing scale, this isn’t just a tool upgrade—it’s a traffic strategy transformation.
As we all know there is a holocaust going on in Gaza. What are some things that can be done technologically to fight against this.