Context is king: How Avride uses cloud VLMs as a safety net for delivery robots
Avride has integrated vision-language models into its delivery robots. Source: Avride Avride Inc. has built its delivery robots for high level of autonomy. Every single day, hundreds of them navigate busy city streets entirely on their own, processing complex sensor data locally on their onboard compute units. Our sidewalk robots run with minimal human involvement, reliably handling standard urban maneuvers, pedestrians, and traffic lights on their own. However, efficiently managing the mechanic
Delivery robots that operate autonomously on city streets use onboard sensors and local neural networks to handle basic navigation and object detection, but they struggle to understand complex real-world context, such as distinguishing a police officer on routine business from an active crime scene. To address this gap, one company has integrated cloud-based vision-language models that analyze camera snapshots every few seconds to detect unusual or sensitive situations and alert human remote assistants when intervention may be needed. The system preserves privacy by anonymizing faces and license plates on the robot before data leaves the device, and the company uses a plug-and-play architecture to test multiple state-of-the-art models rather than relying on a single provider. This approach evolved from the company's existing practice of using cloud VLMs to automatically filter video data for rare, valuable training scenarios, a technique that proved effective enough to extend into live safety operations.
One short email a day with the stories that matter. Free. Unsubscribe with one click.
Want to go deeper than the news? Explore live, cohort-based AI courses taught by practitioners.
Browse AI courses on Maven
Google DeepMind releases new advanced robotics AI model and next-generation gene prediction model
Alphabet Inc.’s artificial intelligence research lab, Google DeepMind, Tuesday debuted the next generation of Gemini Robotics, the company’s most capable vision-language-action model to date for creating an “AI brain” for robotics. Separately, DeepMind announced AlphaGenome 2 today, a next-generation AI model that can predict how DNA sequences and their variations affect human cell types, creating […] The post Google DeepMind releases new advanced robotics AI model and next-generation gene
Also reported by MarkTechPost
Gemini Robotics 2.5 accelerates real-world general dexterity
Our most capable vision-language-action model enhances robot dexterity, reliability, and safety
Also reported by Ars Technica
Can AI agents program industrial robots without training data? Berkeley says yes
The new RoboSimStudio is an open-source project that enables agentic AI coding of robotic simulation environments. | Credit: UC Berkeley UC Berkeley and NVIDIA have released open-source tools for a new approach: enabling teams of AI coding agents to write, simulate, and refine readable robot programs. Since the early days of industrial robots, programming, tuning, and troubleshooting robotic workcells has taken weeks to months. New research coming out of UC Berkeley has demonstrated that new age
Boston Dynamics adds wheels to Atlas humanoid to test whole-body control
Bridging the divide between deterministic physics-based control and policy-driven adaptation, Boston Dynamics has deployed a hybrid locomotion architecture on a wheeled variant of its Atlas humanoid. The “experiment” combined classical model-based motor control with whole-body reinforcement learning to maintain dynamic stability across unstructured, real-world terrain. In the video, Andy Wu, controls engineer for Atlas at Boston Dynamics, highlights some recent experiments that he
Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in Simulation
Simulation, which has long been the proving ground for robots learning to walk, roll and hop, is now helping researchers teach robots tasks that require a finer touch: aligning gears, connecting cables and assembling small components. The shift from locomotion to manipulation is taking center stage at this week’s Conference on Robot Learning (CoRL) in Austin, Texas, alongside the rise of agentic robotics — a new paradigm that integrates frontier agents to create interpretable code for physical
Memories.ai builds standardized egocentric video pipeline
Memories.ai models label every clip: Memories-H0 for hand pose, EgoAction for action segments. | Credit: Memories.ai Memories.ai is building a data pipeline for robotics companies that combines standardized video collection with AI-powered annotation to turn everyday egocentric footage into training data for robots. The San Francisco-based company is developing software and AI to help robots understand the physical world. It builds video-understanding models and data infrastructure, and it colle