📌 TOPINDIATOURS Breaking ai: ByteDance Introduces Astra: A Dual-Model Architecture

The increasing integration of robots across various sectors, from industrial manufacturing to daily life, highlights a growing need for advanced navigation systems. However, contemporary robot navigation systems face significant challenges in diverse and complex indoor environments, exposing the limitations of traditional approaches. Addressing the fundamental questions of “Where am I?”, “Where am I going?”, and “How do I get there?”, ByteDance has developed Astra, an innovative dual-model architecture designed to overcome these traditional navigation bottlenecks and enable general-purpose mobile robots.

Traditional navigation systems typically consist of multiple, smaller, and often rule-based modules to handle the core challenges of target localization, self-localization, and path planning. Target localization involves understanding natural language or image cues to pinpoint a destination on a map. Self-localization requires a robot to determine its precise position within a map, especially challenging in repetitive environments like warehouses where traditional methods often rely on artificial landmarks (e.g., QR codes). Path planning further divides into global planning for rough route generation and local planning for real-time obstacle avoidance and reaching intermediate waypoints.

While foundation models have shown promise in integrating smaller models to tackle broader tasks, the optimal number of models and their effective integration for comprehensive navigation remained an open question.

ByteDance’s Astra, detailed in their paper “Astra: Toward General-Purpose Mobile Robots via Hierarchical Multimodal Learning” (website: https://astra-mobility.github.io/), addresses these limitations. Following the System 1/System 2 paradigm, Astra features two primary sub-models: Astra-Global and Astra-Local. Astra-Global handles low-frequency tasks like target and self-localization, while Astra-Local manages high-frequency tasks such as local path planning and odometry estimation. This architecture promises to revolutionize how robots navigate complex indoor spaces.

Astra-Global: The Intelligent Brain for Global Localization

Astra-Global serves as the intelligent core of the Astra architecture, responsible for critical low-frequency tasks: self-localization and target localization. It functions as a Multimodal Large Language Model (MLLM), adept at processing both visual and linguistic inputs to achieve precise global positioning within a map. Its strength lies in utilizing a hybrid topological-semantic graph as contextual input, allowing the model to accurately locate positions based on query images or text prompts.

The construction of this robust localization system begins with offline mapping. The research team developed an offline method to build a hybrid topological-semantic graph G=(V,E,L):

• V (Nodes): Keyframes, obtained by temporal downsampling of input video and SfM-estimated 6-Degrees-of-Freedom (DoF) camera poses, act as nodes encoding camera poses and landmark references.
• E (Edges): Undirected edges establish connectivity based on relative node poses, crucial for global path planning.
• L (Landmarks): Semantic landmark information is extracted by Astra-Global from visual data at each node, enriching the map’s semantic understanding. These landmarks store semantic attributes and are connected to multiple nodes via co-visibility relationships.

In practical localization, Astra-Global’s self-localization and target localization capabilities leverage a coarse-to-fine two-stage process for visual-language localization. The coarse stage analyzes input images and localization prompts, detects landmarks, establishes correspondence with a pre-built landmark map, and filters candidates based on visual consistency. The fine stage then uses the query image and coarse output to sample reference map nodes from the offline map, comparing their visual and positional information to directly output the predicted pose.

For language-based target localization, the model interprets natural language instructions, identifies relevant landmarks using their functional descriptions within the map, and then leverages landmark-to-node association mechanisms to locate relevant nodes, retrieving target images and 6-DoF poses.

To empower Astra-Global with robust localization abilities, the team employed a meticulous training methodology. Using Qwen2.5-VL as the backbone, they combined Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO). SFT involved diverse datasets for various tasks, including coarse and fine localization, co-visibility detection, and motion trend estimation. In the GRPO phase, a rule-based reward function (including format, landmark extraction, map matching, and extra landmark rewards) was used to train for visual-language localization. Experiments showed GRPO significantly improved Astra-Global’s zero-shot generalization, achieving 99.9% localization accuracy in unseen home environments, surpassing SFT-only methods.

Astra-Local: The Intelligent Assistant for Local Planning

Astra-Local acts as the intelligent assistant for Astra’s high-frequency tasks, a multi-task network capable of efficiently generating local paths and accurately estimating odometry from sensor data. Its architecture comprises three core components: a 4D spatio-temporal encoder, a planning head, and an odometry head.

The 4D spatio-temporal encoder replaces traditional mobile stack perception and prediction modules. It begins with a 3D spatial encoder that processes N omnidirectional images through a Vision Transformer (ViT) and Lift-Splat-Shoot to convert 2D image features into 3D voxel features. This 3D encoder is trained using self-supervised learning via 3D volumetric differentiable neural rendering. The 4D spatio-temporal encoder then builds upon the 3D encoder, taking past voxel features and future timestamps as input to predict future voxel features through ResNet and DiT modules, providing current and future environmental representations for planning and odometry.

The planning head, based on pre-trained 4D features, robot speed, and task information, generates executable trajectories using Transformer-based flow matching. To prevent collisions, the planning head incorporates a masked ESDF loss (Euclidean Signed Distance Field). This loss calculates the ESDF of a 3D occupancy map and applies a 2D ground truth trajectory mask, significantly reducing collision rates. Experiments demonstrate its superior performance in collision rate and overall score on out-of-distribution (OOD) datasets compared to other methods.

The odometry head predicts the robot’s relative pose using current and past 4D features and additional sensor data (e.g., IMU, wheel data). It trains a Transformer model to fuse information from different sensors. Each sensor modality is processed by a specific tokenizer, combined with modality embeddings and temporal positional embeddi…

Konten dipersingkat otomatis.

🔗 Sumber: syncedreview.com

📌 TOPINDIATOURS Update ai: After Outcry, Firefox Promises “Kill Switch” That Turns

The backlash against AI invading almost every aspect of the computing experience is growing by the day.

Particularly as an onslaught of lazy AI slop subsuming news feeds, the tech is starting to feel like a massive distraction — and huge parts of the internet are disillusioned or even fuming in anger.

For instance, a vast number of Windows users refused to upgrade after Microsoft announced it would turn the operating system into a so-called “agentic OS.”

Even household names in the open-source industry aren’t safe. After being appointed as the new CEO of open-source software company Mozilla, whose Firefox browser has long been lauded as a compelling alternative to Google’s Chrome and Apple’s Safari, Anthony Enzor-DeMeo announced that it would be tripling down on AI.

In a December 16 blog post, Enzor-DeMeo announced that Firefox would become a “modern AI browser and support a portfolio of new and trusted software additions.”

But a ringing backlash quickly forced the company into damage control mode.

“I’ve never seen a company so astoundingly out of touch with the people who want to use its software,” one disillusioned user tweeted in response to the news.

“I switched back to Firefox late last year BECAUSE it was the last AI-free browser,” another lamented. “I shoulda known.”

“Please don’t turn Firefox into an AI browser,” yet another begged. “That’s a great way to push us to alternatives.”

The outcry was formidable enough for Mozilla to clarify the company’s new CEO’s comments.

“Something that hasn’t been made clear: Firefox will have an option to completely disable all AI features,” the company wrote in an update on Mastodon. “We’ve been calling it the AI kill switch internally. I’m sure it’ll ship with a less murderous name, but that’s how seriously and absolutely we’re taking this.”

An open letter posted to the Firefox subreddit took issue with Enzor-DeMeo’s new direction.

“Ironically, in a post announcing this new direction and highlighting ‘agency and choice,’ there was little mention of user input or feedback,” the letter reads. “This highlights a disconnect that many of us experience daily: Mozilla has a pattern of struggling to implement and support basic features, and much of the time fails to even acknowledge serious user feedback.”

“Firefox doesn’t need to become Google or Microsoft to succeed by both business and user standards,” the letter goes on. “It’s beloved precisely because it’s not. I hope that distinction isn’t lost as Mozilla enters its ‘next chapter’ as part of a ‘broader ecosystem of trusted software.’”

In an apparent effort to reassure the company’s most diehard fans, Enzor-DeMeo took to the comments.

“Rest assured, Firefox will always remain a browser built around user control,” he wrote. “That includes AI. You will have a clear way to turn AI features off. A real kill switch is coming in Q1 of 2026.”

However, his attempts to calm the situation ended up fanning the flames even further.

“If a ‘kill switch’ is the official control for this, then the entire organization needs to stop referring to your ‘AI’ features as ‘opt-in,’” one user responded. “This is clearly opt-out.”

“If Mozilla can’t agree to that basic definition, I don’t see how users are supposed to trust it’ll actually work,” the user added.

Interestingly, the competing browser company Vivaldi, whose browser is based on Google’s open-source Chromium project, has taken a dramatically different approach.

In an August blog post, Vivaldi CEO Jon von Tetzchner accused other companies like Google and Microsoft of “reshaping the address bar into an assistant prompt, turning the joy of exploring into inactive spectatorship.”

“We will continue building a browser for curious minds, power users, researchers, and anyone who values autonomy,” von Tetzchner wrote.

“If AI contributes to that goal without stealing intellectual property, compromising privacy or the open web, we will use it,” he added. “If it turns people into passive consumers, we will not.”

More on AI slop software: Vast Number of Windows Users Refusing to Upgrade After Microsoft’s Embrace of AI Slop

The post After Outcry, Firefox Promises “Kill Switch” That Turns Off All AI Features appeared first on Futurism.

🔗 Sumber: futurism.com