MAROKO133 Breaking ai: ByteDance Introduces Astra: A Dual-Model Architecture for Autonomou

📌 MAROKO133 Hot ai: ByteDance Introduces Astra: A Dual-Model Architecture for Auto

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

📌 MAROKO133 Hot ai: CEO of Roblox Says Child Predators on the Platform Are an “Opp

Though the children’s gaming platform Roblox has notched over 200 million daily users at its peak, they’re not all there for fun and games. Active since 2006, the extremely popular online game has become a haven for bad actors, like racketeers running online casinos for underaged users, or pedophiles prowling for victims.

It’s an issue Roblox has been keenly aware of, as 20 federal lawsuits against the company bring children’s safety on the platform to national attention. In response to the intense backlash, the company has rolled out a facial recognition system meant to gate users of similar ages in with each other. That system opens a whole other can of worms, but now it’s comments made by Roblox CEO and co-founder David Baszucki that are raising the most immediate eyebrows.

In a lengthy new interview on the New York Times‘ “Hard Fork” podcast, Baszucki didn’t downplay the issue of child exploitation as some might expect. Instead, he bafflingly framed it as an “opportunity.“

Asked by co-host Casey Newton how he thinks about the issue of child predators on his platform, the Roblox CEO said the issue wasn’t just a “problem, but an opportunity as well.”

“How do we allow young people to build, communicate and hang out together?” he mused. “How do we build the future of communication at the same time? So we, you know, we’ve been, I think in a good way, working on this ever since we started… and so fast-forward to where we are today, it’s just like every week, what is the latest tech? At the scale we’re at, 150 million daily actives, 11 billion hours a month, like what is the best way to keep pushing this forward?”

Later on, when the hosts asked Baszucki about challenges of handling user safety on such an astronomical scale, the CEO again presented the issue as a business opportunity — specifically, an opportunity to carve a monopoly out the online gaming scene.

“I think we actually see there being an incredible opportunity,” Baszucki replied. “Like, the gaming space, in a way, is $190 billion. So, we now have about three or four percent of that — I guess three percent — coming through Roblox. So I would say we like the scale: it creates an opportunity for individual game creators who might be making their game by themselves, without part of systems like that, to make it as part of a more overall system.”

That “overall system,” of course, is Roblox, with all its baked-in chat features and historically poor user moderation.

The Hard Fork hosts went on to grill Baszucki on a 2024 report by short seller and activist firm Hindenburg Research, titled “Roblox: Inflated Key Metrics For Wall Street And A Pedophile Hellscape For Kids.” As the colorful title outlines, the report’s thesis was that Roblox was lowering its spending on user safety in order to report growth to investors.

“First off, Hindenburg is no longer in existence, correct? So, you should report on that,” Baszucki dodged, referencing the firm’s voluntary closure earlier this year. “They went out of business for some reason.”

“So it’s really interesting, because I think we’re diving into a situation where we’re getting better, better, better,” the Roblox CEO continued. “But would you ask the same situation of someone who converted from maybe hyper-manual labor making cars by hand to an assembly line?”

In other words, Baszucki claims that the reason for the dip in safety spending is directly linked to the adoption of AI systems. One of those moderation systems, announced over the summer, was supposedly designed to detect “early signs” of child endangerment. However, under Baszucki’s social media posts about the tool, you can find dozens of complaints that the AI is failing to stop harmful content.

“MeepCity [a game on Roblox]… has seen extreme amounts of inappropriate content in it for years,” one user warned. “It does not appeared that effective action to stop violative behavior has ever been taken. Is AI moderation incapable of detecting this kind of content or is it being allowed?”

It’s too soon to tell whether the new facial recognition software will be enough to stop sexual predators from abusing the platform — but considering Roblox’s long history of ineffective stop gaps, nevermind its CEO’s combative crash out on Hard Fork, we’re not holding our breaths.

More on content moderation: If You Thought Facebook Was Toxic Already, Now It’s Replacing Its Human Moderators with AI

The post CEO of Roblox Says Child Predators on the Platform Are an “Opportunity” appeared first on Futurism.

🔗 Sumber: futurism.com