


Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o
Deep integration of vision and robot learning.
When two robot hands work together smoothly to fold clothes, pour tea, and pack shoes, plus the 1X humanoid robot NEO that has been making headlines recently, you may There is a feeling that we are beginning to enter the age of robots.

In fact, these silky movements are the product of advanced robotic technology + exquisite frame design + multi-modal large models.
We know that useful robots often require complex and exquisite interactions with the environment, and the environment can be expressed as constraints in the spatial and temporal domains.
For example, if you want a robot to pour tea, the robot first needs to grasp the handle of the teapot and keep it upright without spilling the tea, and then move it smoothly until the mouth of the pot is aligned with the mouth of the cup. Then tilt the teapot at an angle. Here, constraints include not only intermediate goals (such as aligning the mouth of the pot with the mouth of the cup), but also transition states (such as keeping the teapot upright); together they determine the spatial, temporal and other combination requirements of the robot's actions relative to the environment.
However, the real world is complex and how to construct these constraints is a very challenging problem.
Recently, Li Feifei’s team has made a breakthrough in this research direction and proposed ReKep/Relational Keypoint Constraints. Simply put, this method represents the task as a sequence of relationship key points. Moreover, this framework can also be well integrated with large multi-modal models such as GPT-4o. Judging from the demonstration video, this method performs quite well. The team has also released relevant code. This article is written by Wenlong Huang.

Paper title: ReKep: Spatio-Temporal Reasoning of Relational Keypoint Constraints for Robotic Manipulation
Paper Address: https://rekep-robot.github.io/rekep.pdf
Project website: https://rekep-robot.github.io
Code address: https://github.com/huangwl18/ReKep
Li Feifei said that this work demonstrates a deeper integration of vision and robot learning! Although the paper does not mention World Labs, an AI company focusing on spatial intelligence founded by Li Feifei in early 5 this year, ReKep clearly has great potential in spatial intelligence.

Method

Relationship Key Point Constraint (ReKep)
First, let’s look at a ReKep instance. It is assumed here that a set of K key points has been specified. Specifically, each keypoint k_i ∈ ℝ^3 is a 3D point on the scene surface with Cartesian coordinates.
A ReKep instance is a function like this: ?: ℝ^{K×3}→ℝ; it can map a set of key points (denoted as ?) into an unbounded cost, When ?(?) ≤ 0, the constraint is satisfied. As for the specific implementation, the team implemented the function ? as a stateless Python function that contains NumPy operations on key points, which may be non-linear and non-convex. Essentially, a ReKep instance encodes a desired spatial relationship between keypoints.
However, an operation task usually involves multiple spatial relationships and may have multiple time-related phases, each of which requires different spatial relationships. To this end, the team's approach is to decompose a task into N stages and use ReKep to specify two types of constraints for each stage i ∈ {1, ..., N}:

A set of sub-goal constraints

A set of path constraints
where encodes a key point relationship to be achieved at the end of stage i, and
encodes a key point relationship to be satisfied for each state within stage i. Take the tea pouring task in Figure 2 as an example, which consists of three stages: grabbing, aligning, and pouring tea.
The Phase 1 subgoal constraint is to reach the end effector toward the teapot handle. The stage 2 sub-goal constraint is to keep the mouth of the teapot above the mouth of the cup. Additionally, the stage 2 path constraint is to keep the teapot upright to avoid spilling the tea. The final stage 3 subgoal constraint is to reach the specified tea pouring angle.
Use ReKep to define the operation task as a constrained optimization problem
Use ReKep to convert the robot operation task into a constrained optimization problem involving sub-goals and paths. The end-effector pose is denoted here as ? ∈ SE (3). In order to perform the operation task, the goal here is to obtain the overall discrete-time trajectory?_{1:T}:

That is, for each stage i, the optimization The goal of the problem is: given a given set of ReKep constraints and auxiliary costs, find an end-effector pose as the next subgoal (and its associated time), and the pose sequence that achieves this subgoal. This formula can be considered as direct shooting in trajectory optimization.
Decomposition and algorithm instantiation
In order to solve the above formula 1 in real time, the team chose to decompose the overall problem and only focus on the next sub-goal and achieve the The corresponding path of the sub-goal is optimized. Algorithm 1 gives the pseudocode of this process.

The solution formula for the sub-goal problem is:

The solution formula for the path problem is:

Backtracking
The real environment is complex and changeable. Sometimes during the task, the sub-goal constraints of the previous stage may no longer hold (for example, when the tea cup is poured) taken away), this time you need to re-plan. The team's approach is to check the path for problems. If problems are found, iteratively go back to the previous stage.

Forward model of key points
In order to solve equations 2 and 3, the team used a forward model h, which can be used during the optimization process Estimate Δ? based on Δ?. Specifically, given a change in the end-effector pose Δ?, the change in key point position is calculated by applying the same relative rigidity transformation ?′[grasped] = T_{Δ?}・?[grasped], while assuming other key The point remains still.
Key point proposal and ReKep generation
In order to allow the system to freely perform various tasks in real situations, the team also used large models! Specifically, they designed a pipeline process for keypoint proposal and ReKep generation using a large visual model and a visual-linguistic model.
Key point proposal
Given an RGB image, DINOv2 is first used to extract patch-level features F_patch. Bilinear interpolation is then performed to upsample the features to the original image size, F_interp. To ensure that the proposals cover all relevant objects in the scene, they used Segment Anything (SAM) to extract all masks M = {m_1, m_2, ... , m_n} in the scene.
For each mask j, cluster the mask features F_interp[m_j] using k-means (k = 5) and cosine similarity measures. The centroids of the clusters are used as candidate keypoints, which are then projected to world coordinates ℝ^3 using a calibrated RGB-D camera. Other candidates within 8cm of the candidate keypoint will be filtered out. Overall, the team found that this process can identify a large number of fine-grained and semantically meaningful object regions.
ReKep generation
After obtaining the candidate key points, they are then superimposed on the original RGB image and labeled with numbers. Combined with the language instructions of the specific task, GPT-4o is then queried to generate the number of required stages and the sub-goal constraints and path constraints corresponding to each stage i.
Experiment
The team verified the constraint design through experiments and tried to answer the following three questions:
1. The How well does the framework's automated build and composition operations behave?
2. システムは新しいオブジェクトや操作戦略にどの程度うまく一般化できますか?
3. さまざまなコンポーネントがシステム障害にどのような影響を与える可能性がありますか?
ReKep を使用して 2 つのロボット アームを操作
彼らは、多段階 (m)、現場/実践シナリオ (w)、両手操作 (b) とリアクション (r) の動作。これらのタスクには、お茶を注ぐ (m、w、r)、本を並べる (w)、缶をリサイクルする (w)、箱にテープを貼る (w、r)、洗濯物をたたむ (b)、靴を梱包する (b)、共同で折りたたむ (b、 r)。
結果を表 1 に示し、成功率データを報告します。

全体として、新しく提案されたシステムは、タスク固有のデータや環境モデルが提供されない場合でも、正しい制約を構築し、非構造化環境で実行できます。特に、ReKep は各タスクの中核となるパズルを効果的に処理します。
実際の実行プロセスのアニメーションをいくつか示します:

操作戦略の一般化
チームは、洗濯物をたたむタスクに基づいて、新しい戦略の一般化パフォーマンスを調査しました。端的に言えば、システムがさまざまな種類の衣類を折りたためるかどうかを確認することです。これには、幾何学と常識的な推論が必要です。ここでは

GPT-4o が使用されており、プロンプトにはコンテキスト例のない一般的な手順のみが含まれています。 「戦略の成功」は、生成された ReKep が実行可能であることを意味し、「実行の成功」は、衣類の種類ごとに、特定の実行可能な戦略のシステム成功率を測定します。
結果は興味深いものです。システムは衣服ごとに異なる戦略を使用しており、衣服をたたむ方法の一部は人間が一般的に使用するものと同じであることがわかります。


システムエラーの分析
フレームワークの設計はモジュール式であるため、簡単にシステムエラーの解析に便利です。チームは、表 1 の実験で発生した障害ケースを手動で検査し、これに基づいて、パイプライン プロセスにおける時間的な依存関係を考慮して、モジュールがエラーを引き起こす可能性を計算しました。結果を図 5 に示します。

さまざまなモジュールの中で、キー ポイント トラッカーが最も多くのエラーを生成することがわかります。これは、頻繁かつ断続的なオクルージョンによってシステムが正確に追跡することが困難になるためです。
The above is the detailed content of Li Feifei's team proposed ReKep to give robots spatial intelligence and integrate GPT-4o. For more information, please follow other related articles on the PHP Chinese website!

Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

Video Face Swap
Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Article

Hot Tools

Notepad++7.3.1
Easy-to-use and free code editor

SublimeText3 Chinese version
Chinese version, very easy to use

Zend Studio 13.0.1
Powerful PHP integrated development environment

Dreamweaver CS6
Visual web development tools

SublimeText3 Mac version
God-level code editing software (SublimeText3)

Hot Topics

But maybe he can’t defeat the old man in the park? The Paris Olympic Games are in full swing, and table tennis has attracted much attention. At the same time, robots have also made new breakthroughs in playing table tennis. Just now, DeepMind proposed the first learning robot agent that can reach the level of human amateur players in competitive table tennis. Paper address: https://arxiv.org/pdf/2408.03906 How good is the DeepMind robot at playing table tennis? Probably on par with human amateur players: both forehand and backhand: the opponent uses a variety of playing styles, and the robot can also withstand: receiving serves with different spins: However, the intensity of the game does not seem to be as intense as the old man in the park. For robots, table tennis

On August 21, the 2024 World Robot Conference was grandly held in Beijing. SenseTime's home robot brand "Yuanluobot SenseRobot" has unveiled its entire family of products, and recently released the Yuanluobot AI chess-playing robot - Chess Professional Edition (hereinafter referred to as "Yuanluobot SenseRobot"), becoming the world's first A chess robot for the home. As the third chess-playing robot product of Yuanluobo, the new Guoxiang robot has undergone a large number of special technical upgrades and innovations in AI and engineering machinery. For the first time, it has realized the ability to pick up three-dimensional chess pieces through mechanical claws on a home robot, and perform human-machine Functions such as chess playing, everyone playing chess, notation review, etc.

The start of school is about to begin, and it’s not just the students who are about to start the new semester who should take care of themselves, but also the large AI models. Some time ago, Reddit was filled with netizens complaining that Claude was getting lazy. "Its level has dropped a lot, it often pauses, and even the output becomes very short. In the first week of release, it could translate a full 4-page document at once, but now it can't even output half a page!" https:// www.reddit.com/r/ClaudeAI/comments/1by8rw8/something_just_feels_wrong_with_claude_in_the/ in a post titled "Totally disappointed with Claude", full of

At the World Robot Conference being held in Beijing, the display of humanoid robots has become the absolute focus of the scene. At the Stardust Intelligent booth, the AI robot assistant S1 performed three major performances of dulcimer, martial arts, and calligraphy in one exhibition area, capable of both literary and martial arts. , attracted a large number of professional audiences and media. The elegant playing on the elastic strings allows the S1 to demonstrate fine operation and absolute control with speed, strength and precision. CCTV News conducted a special report on the imitation learning and intelligent control behind "Calligraphy". Company founder Lai Jie explained that behind the silky movements, the hardware side pursues the best force control and the most human-like body indicators (speed, load) etc.), but on the AI side, the real movement data of people is collected, allowing the robot to become stronger when it encounters a strong situation and learn to evolve quickly. And agile

At this ACL conference, contributors have gained a lot. The six-day ACL2024 is being held in Bangkok, Thailand. ACL is the top international conference in the field of computational linguistics and natural language processing. It is organized by the International Association for Computational Linguistics and is held annually. ACL has always ranked first in academic influence in the field of NLP, and it is also a CCF-A recommended conference. This year's ACL conference is the 62nd and has received more than 400 cutting-edge works in the field of NLP. Yesterday afternoon, the conference announced the best paper and other awards. This time, there are 7 Best Paper Awards (two unpublished), 1 Best Theme Paper Award, and 35 Outstanding Paper Awards. The conference also awarded 3 Resource Paper Awards (ResourceAward) and Social Impact Award (

Deep integration of vision and robot learning. When two robot hands work together smoothly to fold clothes, pour tea, and pack shoes, coupled with the 1X humanoid robot NEO that has been making headlines recently, you may have a feeling: we seem to be entering the age of robots. In fact, these silky movements are the product of advanced robotic technology + exquisite frame design + multi-modal large models. We know that useful robots often require complex and exquisite interactions with the environment, and the environment can be represented as constraints in the spatial and temporal domains. For example, if you want a robot to pour tea, the robot first needs to grasp the handle of the teapot and keep it upright without spilling the tea, then move it smoothly until the mouth of the pot is aligned with the mouth of the cup, and then tilt the teapot at a certain angle. . this

This afternoon, Hongmeng Zhixing officially welcomed new brands and new cars. On August 6, Huawei held the Hongmeng Smart Xingxing S9 and Huawei full-scenario new product launch conference, bringing the panoramic smart flagship sedan Xiangjie S9, the new M7Pro and Huawei novaFlip, MatePad Pro 12.2 inches, the new MatePad Air, Huawei Bisheng With many new all-scenario smart products including the laser printer X1 series, FreeBuds6i, WATCHFIT3 and smart screen S5Pro, from smart travel, smart office to smart wear, Huawei continues to build a full-scenario smart ecosystem to bring consumers a smart experience of the Internet of Everything. Hongmeng Zhixing: In-depth empowerment to promote the upgrading of the smart car industry Huawei joins hands with Chinese automotive industry partners to provide

Conference Introduction With the rapid development of science and technology, artificial intelligence has become an important force in promoting social progress. In this era, we are fortunate to witness and participate in the innovation and application of Distributed Artificial Intelligence (DAI). Distributed artificial intelligence is an important branch of the field of artificial intelligence, which has attracted more and more attention in recent years. Agents based on large language models (LLM) have suddenly emerged. By combining the powerful language understanding and generation capabilities of large models, they have shown great potential in natural language interaction, knowledge reasoning, task planning, etc. AIAgent is taking over the big language model and has become a hot topic in the current AI circle. Au
