MAROKO133 Update ai: ByteDance Introduces Astra: A Dual-Model Architecture for Autonomous

📌 MAROKO133 Eksklusif ai: ByteDance Introduces Astra: A Dual-Model Architecture fo

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: E-waste, circular economy, and the limits of regulation Terbar

In this episode of Lexicon, we sit down with Scott Butler, Executive Director of Material Focus, about the growing challenge of electronic waste and why the materials locked inside everyday devices are becoming increasingly critical to our technological, environmental, and geopolitical future.

But before we get into the weeds, remember to subscribe to IE+ for premium insights and exclusive content!

The age of the “urban mine”

When people hear the term e-waste, Butler explains, they often picture “old phones and laptops.” But that, he believes, undersells the problem. 

“When you think about electricals,” he explained, “it encompasses so many things—from American double-door fridge freezers, through to televisions, through to vapes, even as well.”

“All of that is full of really valuable materials,” Butler told us. “We’re talking steel, aluminium, copper, lithium,” he added.

In more advanced electronics, that list expands to include critical technology metals that are now drawing intense geopolitical attention. Yet too often, these materials are treated as disposable. 

“We don’t necessarily value the materials that make them up when we no longer need them,” he added. And this is where the concept of the “urban mine” becomes important.

As Butler explained, rather than viewing old electronics as waste, he argues we should see them as a stockpile of already-extracted materials sitting in homes, offices, and storage spaces. “We have this urban mine of already mined materials that is sitting around,” he said.

Except, this mine isn’t hidden underground; it’s in drawers, cupboards, and garages. Butler noted that “we’ve all got drawers full of random old tech… cables for products we lost a long time ago, DVD remote controls for DVD players that have long gone, and batteries.” 

Individually, those items don’t feel valuable, but collectively they represent a massive untapped resource. “All of that combined is a rich source of material… and that’s the missed opportunity,” he told us.

Copper, in particular, plays a central role. As societies push toward electrification, Butler describes copper as “the silent gold when it comes to green and smart tech futures.” Even old, unused cables suddenly matter when every system—from transport to energy—relies on electricity.

The hidden risks of “fast tech”

Beyond lost materials, Butler highlights a newer and more dangerous trend: fast tech. Similar to fast fashion, fast tech refers to cheap, low-quality electronics designed for short lifespans. 

“We have this phrase fast tech,” Butler explained, describing “this growing cheap technology that often doesn’t work for very long and people throw it away.”

Disposable vapes are a prime example. Designed for convenience, they combine lithium-ion batteries, electronics, plastics, and chemicals in a product meant to be discarded after minimal use. 

“You’ve basically got hundreds of millions of… toxic fire sticks distributed randomly across the UK,” Butler said, “and then someone’s like, well, who’s going to deal with that?”

Improper disposal carries real risks. “Lithium-ion is a wonderful technology,” Butler added, “but it’s also a technology with a temper if you do the wrong thing to it.” When batteries are crushed in bins or waste trucks, fires can result. 

“We’ve seen over 1,200 battery-related fires in the UK in the last year or so—and we think that’s an under-report,” he explained.

These fires don’t just damage infrastructure; they pose health and environmental risks. 

Why recycling isn’t always possible

Despite these risks, Butler is careful not to blame consumers. “The vast majority of people are pro-recycling,” he said. “But we’re just not aware that these electricals should be or can be recycled.”

Some barriers are surprisingly basic. “People aren’t aware sometimes that things are even electrical,” Butler explained, citing examples like electric toothbrushes or disposable vapes. Others are more psychological: time pressure, confusion, and mistrust. 

“There’s no good trying to get people to change their behaviour if what you’re asking them to do is too hard,” he told us. “It’s basic psychology.”

To this end, Material Focus tackled this by first making recycling easier before pushing awareness. “We focused on building that locator,” Butler told us, referring to a postcode-based tool that helps people find nearby drop-off points.

“Today, there are over 30,000 electrical recycling locations across the UK,” he added.

But convenience alone isn’t enough. Butler stresses the importance of trust and tone. “We’re not preachy. We’re very positive,” he said. 

That philosophy led to the campaign’s unusual mascot: HypnoCat. “A techno-loving, hypnotizing, mesmerizing cat who comes out of TV screens to get people leaning in,” he explained. 

In a crowded media landscape competing with “Taylor Swift, Netflix, and Disney,” he added, environmental messaging has to work harder to earn attention.

Recycling, reuse, and the limits of the circular economy

While recycling is essential, Butler is clear that it’s not the only, or indeed always the best, solution. “Keeping a working, useful product in use for longer is far better than replacing it with a new one,” he told us, because most environmental impact happens during design, manufacturing, and shipping.

However, the picture isn’t simple. For major appliances like washing machines, improvements in energy and water efficiency may eventually outweigh the benefits of keeping older models running. 

Material Focus is now working with University College London to identify where that tipping point lies. Another complicating factor is software. 

“You can have a piece of hardware that’s working fine,” Butler said, “but software changes suddenly render it insecure or obsolete.” 

He points to examples like the end of support for older operating systems, pushing otherwise functional devices toward disposal.

Policy, regulation, and reality

Throughout the discussion, Butler emphasized the gap between policy ambition and real-world systems. “Circular economy poses a very attractive vision,” he added, “but we need the pragmatism and realism about how we get there.” 

Waste systems are deeply embedded in local culture, infrastructure, and regulation, making rapid change difficult.

“There’s a lot of criticism put on the waste electrical regulations,” Butler noted, but he cautions against expecting any single policy to solve everything. 

“Not one single piece of regulation can fix that circular challenge,” he added.

He also highlights unintended consequences, such as regulations that penalize companies experimenting with leasing or service-based …

Konten dipersingkat otomatis.

🔗 Sumber: interestingengineering.com