Boston Dynamics has unveiled a fully electric successor to its hydraulic Atlas robot, designed for real-world commercial applications. The new electric Atlas boasts increased strength, a broader range of motion, and enhanced manipulation capabilities with various gripper options to handle diverse objects in customer environments.
A capable hardware platform is just one part of a successful commercial robotics solution. Boston Dynamics emphasizes the importance of an integrated digital ecosystem - IT infrastructure, workflow integration, safety standards, operational processes, and data management. Their Orbit software provides a centralized fleet management system, with Atlas integration planned alongside Spot and Stretch.
Significant progress has been made on the software front, with AI/machine learning advancements like reinforcement learning and computer vision enabling adaptation to complex situations. Model predictive control and simulation expertise are leveraged for robust operation.
The humanoid form factor allows Atlas to navigate human-designed spaces while exceeding human mobility constraints. Boston Dynamics aims to equip Atlas with efficient whole-body motion capabilities optimized for task completion rather than mimicking human ranges.
Under the hood, the electric Atlas combines decades of innovation in areas like dynamic balancing, bimanual manipulation, and pushing mobility limits through R&D platforms like PETMAN and parkour demonstrations.
With commercial deployments of Spot and Stretch, Boston Dynamics has gained experience translating advanced robotics into real-world value. Strategic partnerships like Hyundai will enable iterative development and testing of Atlas applications.
Read More: An Electric New Era for Atlas | Boston Dynamics
The blog identifies four innovative use cases envisioned for voice AI assistants or voicebots powered by large language models and natural language processing capabilities:
Personal Fashion Stylist A smartphone-based voicebot acts as a personalized fashion consultant. Through conversational inputs, it captures the user's style preferences, body type, and existing wardrobe to recommend tailored outfit combinations suiting their unique needs.
Culinary Assistant
This voicebot serves as an interactive cooking guide, providing personalized recipe suggestions based on available ingredients, dietary restrictions, and custom requests. It offers step-by-step voice instructions to guide users through recipes, reducing kitchen errors.Interactive Study Companion Designed for students, this voicebot takes the role of a study partner. It can quiz learners on key concepts, elaborate on complex topics, and provide memorization tips - leveraging its ability to answer follow-up questions and generate explanatory content.
Virtual Fitness Coach Acting as a voice AI personal trainer, this assistant creates customized workout plans aligning with the user's fitness goals. It tracks workout progress, provides real-time exercise feedback, and aims to boost motivation through encouraging dialogue.
The common thread is utilizing voice user interfaces coupled with large language models to offer personalized, context-aware recommendations and guidance across various lifestyle domains like fashion, cooking, education, and fitness - capitalizing on natural language interactions and generation capabilities.
Read more : Vapi - Voice Assistant
Now with bigger context windows and a refreshed chat UI
The blog announces version 1.14.0 of Cody for VS Code, which includes several enhancements focused on improving the integration of large language models (LLMs) like Claude and GPT-4 for coding assistance.
A key update is expanded context window sizes when using Claude 3's LLMs like Sonnet and Opus models. The maximum context that can be passed to these models has been increased substantially:
30,000 tokens for user-defined context files
15,000 tokens for continuous conversation context
This leverages Claude 3's improved context recall capabilities, allowing much more code and conversational history to be provided as context for better answer quality from the LLM.
The output token limit for LLM-generated responses has also been quadrupled from 1,000 to 4,000 tokens across all models and pricing tiers.
On the UI front, the chat interface has been revamped to show a cell-based structure displaying context fetched by Sourcegraph alongside user and LLM messages.
For JavaScript, TypeScript, Go and Python, Cody's unit test generation command now determines the relevant code range automatically and offers a quick action option when triggered on function symbols.
Users can now easily add highlighted code ranges from open files as context mentions (@-mentions) in the chat sidebar via right-click.
In terms of LLM/GenAI model support, the new Mixtral 8x22B open source model has been added as an option for Cody Pro users, lauded for its performance on coding and math benchmarks among open models. The GPT-4 Turbo model has also been upgraded to the newer non-preview version.
Experimental integration with Google's CodeGemma models via Ollama is also included, enabling code completion capabilities powered by GenAI models focused on programming tasks.
Overall, this release aims to enhance the coding assistance experience by improving context handling, UI integration, and expanding the range of large language models available within the Cody environment.
Read More : Sourcegraph
