Stanford Seminar - Why would we want a multi-agent system unstable

Stanford Online · Beginner ·🤖 AI Agents & Automation ·3y ago
Mrdjan Jankovic of Ford Research January 13, 2023 In everyday driving, many traffic maneuvers such as merges, lane changes, passing through an intersection, require negotiation between independent actors/agents. The same is true for mobile robots autonomously operating in a space open to other agents (e.g., humans, robots, etc.). Negotiation is an inherently difficult concept to code into a software algorithm. It has been observed in computer simulations that some “decentralized” algorithms produce gridlocks while others never do. It has turned out that gridlocking algorithms create locally stable equilibria in the joint inter-agent space, while, for those that don’t gridlock, equilibria are unstable – hence the title of the talk. We use Control Barrier Function (CBF) based methods to provide collision avoidance guarantees. The main advantage of CBFs is that they result in relatively easier to solve convex programs even for nonlinear system dynamics and inherently non-convex obstacle avoidance problems. Six different CBF-based control policies were compared for collision avoidance and liveness (fluidity of motion, absence of gridlocks) on a 5-agent, holonomic-robot system. The outcome was then correlated with stability analysis on a simpler, yet representative problem. The results are illustrated by extensive simulations and a vehicle experiment with stationary obstacles. 0:00 Introduction 0:49 Objective - unstable feedback loop? ord 2:43 Why CBFs? Short answer - convex QP 6:31 CBF based safety filter 7:58 Barrier margin for robustness 9:14 Robust Control Barrier Functions 10:32 Turning obstacles into barriers 12:16 CBF based obstacle avoidance 13:42 Traffic flow and gridlocks 14:42 Avoiding interacting obstacles 17:26 Decentralized multi-agent controllers 19:07 Centralized CBF Controller 20:06 Co-optimization and CCS 23:19 PCCA algorithm guarantees 24:00 5 agents Monte Carlo Simulations 24:52 Comparison of CBF based methods 26:08 Deadlock resolution 26:51 Cause
Watch on YouTube ↗ (saves to browser)
Sign in to unlock AI tutor explanation · ⚡30

Playlist

Uploads from Stanford Online · Stanford Online · 0 of 60

← Previous Next →
1 Statistical Learning: 13.2 Introduction to Multiple Testing and Family Wise Error Rate
Statistical Learning: 13.2 Introduction to Multiple Testing and Family Wise Error Rate
Stanford Online
2 Statistical Learning: 13.1 Introduction to Hypothesis Testing II
Statistical Learning: 13.1 Introduction to Hypothesis Testing II
Stanford Online
3 Statistical Learning: 12.R.3 Hierarchical Clustering
Statistical Learning: 12.R.3 Hierarchical Clustering
Stanford Online
4 Statistical Learning: 12.R.2 K means Clustering
Statistical Learning: 12.R.2 K means Clustering
Stanford Online
5 Statistical Learning: 12.R.1 Principal Components
Statistical Learning: 12.R.1 Principal Components
Stanford Online
6 Statistical Learning: 13.R.1 Bonferroni and Holm II
Statistical Learning: 13.R.1 Bonferroni and Holm II
Stanford Online
7 Statistical Learning: 12.6 Breast Cancer Example
Statistical Learning: 12.6 Breast Cancer Example
Stanford Online
8 Statistical Learning: 12.5 Matrix Completion
Statistical Learning: 12.5 Matrix Completion
Stanford Online
9 Statistical Learning: 12.4 Hierarchical Clustering
Statistical Learning: 12.4 Hierarchical Clustering
Stanford Online
10 Statistical Learning: 12.3 k means Clustering
Statistical Learning: 12.3 k means Clustering
Stanford Online
11 Statistical Learning: 13.1 Introduction to Hypothesis Testing
Statistical Learning: 13.1 Introduction to Hypothesis Testing
Stanford Online
12 Stanford Seminar - Introduction to Web3
Stanford Seminar - Introduction to Web3
Stanford Online
13 Stanford Seminar - Designing Equitable Online Experiences
Stanford Seminar - Designing Equitable Online Experiences
Stanford Online
14 Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 1
Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 1
Stanford Online
15 Stanford Seminar - Perceiving, Understanding, and Interacting through Touch
Stanford Seminar - Perceiving, Understanding, and Interacting through Touch
Stanford Online
16 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 2
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 2
Stanford Online
17 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 3
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 3
Stanford Online
18 Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 4
Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 4
Stanford Online
19 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 5
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 5
Stanford Online
20 Stanford Seminar - Evolution of a Web3 Company
Stanford Seminar - Evolution of a Web3 Company
Stanford Online
21 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 6
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 6
Stanford Online
22 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 7
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 7
Stanford Online
23 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 8
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 8
Stanford Online
24 Stanford Seminar - Designing Human-Centered AI Systems for Human-AI Collaboration
Stanford Seminar - Designing Human-Centered AI Systems for Human-AI Collaboration
Stanford Online
25 The Sh*tFixers: Bob Sutton Interviews David Kelley, Design Thinking Superstar
The Sh*tFixers: Bob Sutton Interviews David Kelley, Design Thinking Superstar
Stanford Online
26 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 9
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 9
Stanford Online
27 Women Rise: Sheri Sheppard
Women Rise: Sheri Sheppard
Stanford Online
28 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 10
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 10
Stanford Online
29 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 11
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 11
Stanford Online
30 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 12
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 12
Stanford Online
31 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 13
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 13
Stanford Online
32 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 14
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 14
Stanford Online
33 Stanford Webinar - Cloud Computing: What’s on the Horizon with Dr. Timothy Chou
Stanford Webinar - Cloud Computing: What’s on the Horizon with Dr. Timothy Chou
Stanford Online
34 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 15
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 15
Stanford Online
35 Stanford Seminar - Multi-Sensory Neural Objects: Modeling, Inference, and Applications in Robotics
Stanford Seminar - Multi-Sensory Neural Objects: Modeling, Inference, and Applications in Robotics
Stanford Online
36 Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 16
Stanford CS330: Deep Multi-task & Meta Learning I 2021 I Lecture 16
Stanford Online
37 Stanford Seminar - Toward Better Human-AI Group Decisions
Stanford Seminar - Toward Better Human-AI Group Decisions
Stanford Online
38 Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 17
Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 17
Stanford Online
39 Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 18
Stanford CS330: Deep Multi-Task & Meta Learning I 2021 I Lecture 18
Stanford Online
40 Stanford Webinar - Web3 Considered: Possible Futures for Decentralization and Digital Ownership
Stanford Webinar - Web3 Considered: Possible Futures for Decentralization and Digital Ownership
Stanford Online
41 Stanford Seminar - Ethics Governance-in-the-Making: Bridging Ethics Work & Governance Menlo Report
Stanford Seminar - Ethics Governance-in-the-Making: Bridging Ethics Work & Governance Menlo Report
Stanford Online
42 Stanford Seminar -  Towards Generalizable Autonomy: Duality of Discovery & Bias
Stanford Seminar - Towards Generalizable Autonomy: Duality of Discovery & Bias
Stanford Online
43 Stanford Seminar - ML Explainability Part 1 I Overview and Motivation for Explainability
Stanford Seminar - ML Explainability Part 1 I Overview and Motivation for Explainability
Stanford Online
44 Stanford Seminar - ML Explainability Part 2 I Inherently Interpretable Models
Stanford Seminar - ML Explainability Part 2 I Inherently Interpretable Models
Stanford Online
45 Stanford Seminar - ML Explainability Part 3 I Post hoc Explanation Methods
Stanford Seminar - ML Explainability Part 3 I Post hoc Explanation Methods
Stanford Online
46 Kratika Gupta talks about Stanford's Product Management Program
Kratika Gupta talks about Stanford's Product Management Program
Stanford Online
47 Stanford Seminar - Making Teamwork an Objective Discipline - Sid Sijbrandij CEO & Chairman of GitLab
Stanford Seminar - Making Teamwork an Objective Discipline - Sid Sijbrandij CEO & Chairman of GitLab
Stanford Online
48 Stanford Seminar - ML Explainability Part 4 I Evaluating Model Interpretations/Explanations
Stanford Seminar - ML Explainability Part 4 I Evaluating Model Interpretations/Explanations
Stanford Online
49 Stanford Seminar - Adaptable Robotic Manipulation Using Tactile Sensors
Stanford Seminar - Adaptable Robotic Manipulation Using Tactile Sensors
Stanford Online
50 Stanford Seminar - ML Explainability Part 5 I Future of Model Understanding
Stanford Seminar - ML Explainability Part 5 I Future of Model Understanding
Stanford Online
51 Meet Joe Lapin, Innovation and Entrepreneurship Program Completer
Meet Joe Lapin, Innovation and Entrepreneurship Program Completer
Stanford Online
52 Stanford Seminar: Social Media Scrutiny of Frontline Professionals & Implications for Accountability
Stanford Seminar: Social Media Scrutiny of Frontline Professionals & Implications for Accountability
Stanford Online
53 Stanford Seminar - Alphy and Alphy Reflect: creating a reflective mirror to advance women
Stanford Seminar - Alphy and Alphy Reflect: creating a reflective mirror to advance women
Stanford Online
54 Stanford Webinar - The Digital Future of Health
Stanford Webinar - The Digital Future of Health
Stanford Online
55 Stanford CS229M - Lecture 1: Overview, supervised learning, empirical risk minimization
Stanford CS229M - Lecture 1: Overview, supervised learning, empirical risk minimization
Stanford Online
56 Stanford CS229M - Lecture 2:  Asymptotic analysis, uniform convergence, Hoeffding inequality
Stanford CS229M - Lecture 2: Asymptotic analysis, uniform convergence, Hoeffding inequality
Stanford Online
57 Stanford CS229M - Lecture 3: Finite hypothesis class, discretizing infinite hypothesis space
Stanford CS229M - Lecture 3: Finite hypothesis class, discretizing infinite hypothesis space
Stanford Online
58 Stanford Seminar - Decentralized Finance (DeFi)
Stanford Seminar - Decentralized Finance (DeFi)
Stanford Online
59 Stanford CS229M - Lecture 4: Advanced concentration inequalities
Stanford CS229M - Lecture 4: Advanced concentration inequalities
Stanford Online
60 Stanford Seminar - Bridging AI & HCI: Incorporating Human Values into the Development of AI Tech
Stanford Seminar - Bridging AI & HCI: Incorporating Human Values into the Development of AI Tech
Stanford Online

Related AI Lessons

AI Basics You MUST Understand Before Building AI Agent
Learn the essential AI basics before building an AI agent to ensure a strong foundation
Dev.to · Yisak Bule
AiFinPay: Autonomous Payments for ruvnet/ruflo
Learn how AiFinPay enables autonomous payments for AI agents in multi-agent swarms, streamlining workflows and monetization
Dev.to AI
AiFinPay: Autonomous Payments for cirosantilli/china-dictatorship
Learn how AiFinPay enables autonomous payments for AI agents with a one-line payment SDK, empowering seamless transactions and promoting awareness on cirosantilli/china-dictatorship platform
Dev.to AI
Shopify Just Gave AI Agents an Operating Manual. We Built One for WooCommerce.
Shopify's SKILL.md file provides instructions for AI agents to behave when shopping on Shopify, and a similar manual can be built for WooCommerce
Dev.to AI

Chapters (18)

Introduction
0:49 Objective - unstable feedback loop? ord
2:43 Why CBFs? Short answer - convex QP
6:31 CBF based safety filter
7:58 Barrier margin for robustness
9:14 Robust Control Barrier Functions
10:32 Turning obstacles into barriers
12:16 CBF based obstacle avoidance
13:42 Traffic flow and gridlocks
14:42 Avoiding interacting obstacles
17:26 Decentralized multi-agent controllers
19:07 Centralized CBF Controller
20:06 Co-optimization and CCS
23:19 PCCA algorithm guarantees
24:00 5 agents Monte Carlo Simulations
24:52 Comparison of CBF based methods
26:08 Deadlock resolution
26:51 Cause
Up next
Defining the agentic AI era
Google for Developers
Watch →