Google DeepMind’s new model, Gemini Robotics, integrates its top large language model, Gemini 2.0, to enhance robot capabilities. This integration allows robots to be more dexterous, follow natural-language commands, and generalize across various tasks. Previously, robots struggled with these abilities. The model’s success stems from leveraging advancements in Gemini 2.0, enabling robots to reason about actions, understand human requests, and communicate effectively. Gemini Robotics also generalizes across different robot types. In demonstrations, robots achieved a 90% success rate in tasks like sorting fruits into containers, folding glasses, and even performing a slam dunk with a toy basketball. These tasks were executed even when objects were moved, showcasing the model’s adaptability. Google DeepMind trained the robots using both simulated and real-world data, including teleoperation and video analysis. The robots were tested on the ASIMOV data set, where Gemini 2.0 Flash and Gemini Robotics models showed strong performance in recognizing safe and unsafe scenarios. Additionally, a constitutional AI mechanism based on Asimov’s laws was implemented to ensure safe robot-human interactions.
Physical Intelligence
@physical_int
·
Feb 26 Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly?We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).
Sergey Levine
@svlevine
·
Feb 26 We made π0 “think harder”: our new Hierarchical Interactive Robot (Hi Robot) method “thinks” through complex tasks and prompts, directing π0 to break up complex tasks into basic steps, handling human feedback, and modifying tasks on the fly.
Princeton Computer Science
@PrincetonCS
·
4h Exploration is a key problem in reinforcement learning.
@ben_eysenbach
and Grace Liu have found an unexpected way to get machines to explore: they gave AI agents a single difficult task and no feedback at all.Learn more: https://bit.ly/3R1a6R0
Wes Roth
@WesRothMoney
·
Mar 6 Engineers at Boston Dynamics have revealed insights on “sequencing,” the first practical task mastered by Atlas, showcasing its ability to plan and execute complex, step-by-step actions autonomously.From parkour moves to practical tasks, Atlas continues to redefine what
UriG
@uri_gadot
·
Mar 10 Our Solution – RL-RC-DoT:
We propose to train a lightweight reinforcement learning (RL) agent that dynamically adjusts Quantization Parameters (QPs) at the macro-block level. This ensures task-relevant regions are prioritized while maintaining overall efficiency.
Kevin
@kevineress
·
5h … Prplxity: FocusRot Apply dynamic allocation between subtasks using techniques from LLMs, switching between rule acquisition, strategy planning & error correct phases. Cluster code injection increases model size by 15%. Socratic validation adds .2-.4s latency per interaction.
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.
