Robotics Research

Foreword
Preface
Organization
Contents
Robot Learning
It's Just Semantics: How to Get Robots to Understand the World the Way We Do
1 Introduction
2 Contextual Scene Understanding
2.1 Spatial Predicates: Object-to-Object Relationships
2.2 Affordances: Object-to-Agent and Scene-to-Agent Relationships
2.3 Challenges and Benefits of Embodiment
3 Summary and Outlook
References
Learning Agile, Vision-Based Drone Flight: From Simulation to Reality
1 Introduction
2 High-Speed Flight in the Wild
3 Acrobatic Flight
4 Autonomous Drone Racing
5 Future Work
5.1 Generalization
5.2 Continual Learning
5.3 Autonomous Drone Racing
References
Continual SLAM: Beyond Lifelong Simultaneous Localization and Mapping Through Continual Learning
1 Introduction
2 Related Work
3 Continual SLAM
4 Technical Approach
5 Experimental Evaluation
5.1 Evaluation of Pose Accuracy of CL-SLAM
5.2 Evaluation of Continual SLAM
6 Conclusion
References
Efficiently Learning Single-Arm Fling Motions to Smooth Garments*-12pt
1 Introduction
2 Related Work
3 Problem Statement
4 Motion Primitives and Pipeline
5 Self-Supervised Learning Pipeline for Flinging
5.1 Learning Procedure of the Fling Action
5.2 Adapting to New Garments
5.3 Execution-Time Stopping Rules
6 Experiments
6.1 Comparison of Learning Methods
6.2 Comparison of Different Trajectory Parameterization Methods
6.3 Results for Different Grasp Points
6.4 Training and Testing Across Different Garments
6.5 Stopping Time During Execution
7 Conclusion
References
Learning Long-Horizon Robot Exploration Strategies for Multi-object Search in Continuous Action Spaces
1 Introduction
2 Related Work
3 Learning Long-Horizon Exploration
3.1 Problem Statement
3.2 Technical Approach
4 Experimental Evaluations
4.1 Experimental Setup
4.2 Simulation Experiments
4.3 Real World Experiments
5 Conclusion
References
Visual Foresight with a Local Dynamics Model*-12pt
1 Introduction
2 Related Work
3 Problem Statement
4 Method
4.1 Structuring the Transition Model
4.2 Local Dynamics Model
4.3 Policy Learning
4.4 Sampling Diverse Actions
5 Experiments
5.1 Accuracy of the Local Dynamics Model
5.2 Policy Learning
5.3 Generalization
6 Limitations and Future Work
7 Conclusion
References
Robot Vision
Monocular Camera and Single-Beam Sonar-Based Underwater Collision-Free Navigation with Domain Randomization*-12pt
1 Introduction
2 Related Work
3 Approach
3.1 Multi-modal Deep Reinforcement Learning Navigation
3.2 Image Depth Prediction Network
3.3 Transferable Model
4 Experimental Results
4.1 Training Experimental Settings
4.2 Performance Comparison with Different Sensor Configurations
4.3 Ablation Study with Transferability Tests
4.4 Performance Demonstration in Real-World Environment
4.5 Action Prediction from Static Underwater Images
5 Conclusion and Future Work
References
Nonmyopic Distilled Data Association Belief Space Planning Under Budget Constraints-12pt
1 Introduction
2 Related Work
3 Background and Notations
3.1 Inference
3.2 DA-BSP
4 Methodology
4.1 Constructing the Belief Tree skeleton
4.2 Nonmyopic Distilled Data Association BSP
4.3 No Budget Constraints in Inference
4.4 Hard Budget Constraints in Inference
4.5 The Cost Function
5 Results
6 Conclusions
References
SCIM: Simultaneous Clustering, Inference, and Mapping for Open-World Semantic Scene Understanding*-12pt
1 Introduction
2 Related Work
3 Method
3.1 Preliminaries: Scene Understanding as a Clustering Problem
3.2 Identifying Novel Categories
4 Experimental Evaluation
4.1 Method Implementation
4.2 Adaptation of Baselines
4.3 Evaluation Metrics
4.4 Results
4.5 Design Choice Verification
5 Discussion and Outlook
References
6N-DoF Pose Tracking for Tensegrity Robots
1 Introduction
2 Related Work
3 Problem Formulation
4 Method and Implementation
4.1 Pose Initialization
4.2 Iterative Pose Tracking
4.3 Physical Constraints
4.4 Adaptive Weights
4.5 Noise Filtering
5 Experiments
5.1 Hardware Setup
5.2 Data Collection
5.3 Evaluation Metric
5.4 Baselines and Ablation Studies
5.5 Quantitative Results
5.6 Qualitative Results
5.7 Running Time
6 Limitations
7 Conclusion
References
Scale-Invariant Fast Functional Registration*-12pt
1 Introduction
2 Related Works
3 Notation
4 Functional Point Cloud Registration
4.1 The Delta-Distance Between Unordered Sets
4.2 Continuous Relaxation of the Delta-Distance
4.3 Functional Point Cloud Registration
5 Experiments and Application
5.1 Implementation Details
5.2 Rationale for Test Algorithms
5.3 Rationale for Experiment Design
5.4 Robustness Against Noise
5.5 Robustness Against Partial Overlap and Varying Densities
5.6 Time Efficiency
5.7 Sensitivity to Initialization
5.8 Real World Application
6 Conclusions
References
Towards Mapping of Underwater Structures by a Team of Autonomous Underwater Vehicles*-12pt
1 Introduction
2 Related Work
3 Proximal-Distal Mapping Framework
3.1 Robust State Estimation
3.2 Photorealistic Reconstruction
3.3 AUV Navigation
4 Conclusions, Lessons Learned, and Future Directions
References
Grasping and Manipulation
Contact-Implicit Planning and Control for Non-prehensile Manipulation Using State-Triggered Constraints*-12pt
1 Introduction
1.1 Contributions
1.2 Related Work
2 Background
2.1 Contact Model
2.2 Contact-Implicit Trajectory Optimization
2.3 Quasi-static Planar Pushing
3 Approach
3.1 State-Triggered Constraints
3.2 Automatic Contact Constraint Activation
3.3 A Reformulation of Coulomb Friction
3.4 Quasi-static Pushing Planning and Execution
4 Experiments
4.1 Dynamic Pushing
4.2 Quasi-static Planar Pushing
5 Conclusion
References
Mechanical Search on Shelves with Efficient Stacking and Destacking of Objects
1 Introduction
2 Related Work
2.1 Mechanical Search
2.2 Stacking and Destacking
2.3 Monte Carlo Tree Search
3 Problem Statement
3.1 Overview
3.2 Problem Setup
3.3 Objective
4 Methods
4.1 Occupancy Distribution
4.2 Stacked Environments
4.3 Action Generation
4.4 Distribution Area Reduction for Stacked Shelves (DARSS)
4.5 Monte Carlo Tree Search for Stacked Shelves (MCTSSS)
5 Experiments
5.1 Baseline and Oracle Policies
5.2 Simulated Experiments
5.3 DARSS Failure Mode
5.4 Ablation Experiments
5.5 Sensitivity Experiments
5.6 Physical Experiments
6 Conclusion
References
Multi-object Grasping in the Plane
1 Introduction
2 Related Work
3 Problem Statement
4 Multi-object Grasps
5 Multi-object Push Grasps
5.1 Two Necessary Conditions
5.2 Multi-object Diameter Function
5.3 Intersection Area
6 Picking System
6.1 Picking Algorithm
6.2 Frictionless Multi-Object Grasp Planner (GP)
7 Experiments
7.1 Simulation and Physical Experiments
7.2 Evaluation of Necessary Conditions in Simulation
7.3 Grasp Planning Experiments in Simulation
7.4 Physical Experiments
8 Discussion and Future Work
References
Parameter Estimation for Deformable Objects in Robotic Manipulation Tasks
1 Introduction
2 Background and Related Works
2.1 Simulation
2.2 Closing the Sim-to-Real Gap
3 Methods
3.1 Modeling
3.2 Nonlinear Programming for Parameter Estimation
4 Results
5 Conclusion and Ongoing Work
References
Place-and-Pick-Based Re-grasping Using Unstable Placement*-12pt
1 Introduction
2 Related Work
2.1 Place-and-Pick-Based Re-grasping
2.2 Re-grasping with Support from the Environment
3 Methodology
3.1 Intermediate Placement for Re-grasping
3.2 Stable and Semi-stable Placement Searching
3.3 Gripper Sliding Planning
3.4 Re-grasp Graph Construction
3.5 Re-grasp Action Motion Planning
4 Experiments
4.1 Re-grasping in Simulation
4.2 Re-grasping in Real World
5 Conclusion
References
Safe, Occlusion-Aware Manipulation for Online Object Reconstruction in Confined Spaces*-12pt
1 Introduction
2 Parameters of Occlusion-Aware Manipulation
3 Related Work
4 Problem Formulation
5 Algorithms and Analysis
5.1 Occlusion Dependency Graph (ODG)
5.2 A Probabilistically Complete Algorithm
5.3 A Monotone Algorithm
6 System Design
7 Experiments
7.1 Experimental Setup
7.2 Results and Analysis
8 Conclusion and Future Work
References
Systems and Design
Cooperation of Assistive Robots to Improve Productivity in the Nursing Care Field*-12pt
1 Introduction
2 CARE -Cooperation of Ai-Robot Enablers-
3 Verification Experiment for CARE in the Living Lab
4 Nursing Care Guidance Using AR Glasses
5 Conclusions
References
Adaptive Radiation Survey Using an Autonomous Robot Executing LiDAR Scans in the Large Hadron Collider*-12pt
1 Introduction
2 Methodology
3 Robotic System
4 Data Acquisition and Processing
5 Online Trajectory Planning and Control
6 Conclusion
References
Reference-Free Learning Bipedal Motor Skills via Assistive Force Curricula
1 Introduction
2 Related Work
2.1 RL-Based Legged Controller
2.2 Assistive Control
2.3 Learning with Recovery
3 Problem Formulation
3.1 Notation
3.2 Problem Statement
4 Learning with Assistive Force
4.1 Assistance Placement
4.2 Assistance Quantity
4.3 Assistance Embedment
4.4 Curriculum Stages
4.5 Learning Settings
5 Experiments
5.1 Example 1: Locomotion
5.2 Example 2: Dynamic Skills
6 Conclusion and Discussion
References
Ball-and-Socket Joint Pose Estimation Using Magnetic Field
1 Introduction
2 Hardware
3 Methodologies
3.1 Problem Statement
3.2 Recurrent Neural Network
3.3 Deep Variational Bayes Filters
4 Experiments
4.1 Experiment Setup
4.2 Experiment Results
5 Conclusions
References
BulletArm: An Open-Source Robotic Manipulation Benchmark and Learning Framework
1 Introduction
2 Related Work
3 Architecture
3.1 Design Philosophy
3.2 Environment
3.3 Expert Demonstrations
4 Environments
4.1 Open-Loop Environments
4.2 Close-Loop Environments
5 Benchmark
5.1 Open-Loop 3D Benchmark
5.2 Close-Loop 4D Benchmark
6 Conclusions
References
Optimization-Based Online Flow Fields Estimation for AUVs Navigationpg*-12pt
1 Introduction
2 Related Works
3 Problem Formulation
4 Flow Field Estimation
4.1 Eddy-Based Constraint
4.2 Incompressibility-Based Constraint
5 Experiment
5.1 Flow Field Estimation
5.2 AUV Navigation
6 Conclusions
References
Flying Hydraulically Amplified Electrostatic Gripper System for Aerial Object Manipulationpg*-12pt
1 Introduction
1.1 Objectives
1.2 Related Work
2 Methods
2.1 HASEL Challenges
2.2 Actuator Manufacturing
2.3 RAPTOR Quadcopter
3 Inverse vs. Classical Gripper Approach
4 Actuator Types and Configurations
4.1 Pouch Width Test (PWT)
4.2 Pouch Length Test (PLT)
4.3 Scorpion Concept Test
4.4 Complete Finger Configuration
5 Drone Integration
5.1 Flight Experiment
6 Conclusion and Future Work
References
Control
Hybrid Quadratic Programming - Pullback Bundle Dynamical Systems Control
1 Introduction
2 Background
3 Method
4 Preliminary Results
5 Conclusion
References
Riemannian Geometry as a Unifying Theory for Robot Motion Learning and Control
1 Motivation
2 Geodesic Synergies: Basics and Proof of Concept
2.1 Riemannian Geometry of Mechanical Systems
2.2 Robot Motion Generation with Geodesic Synergies
3 Vision and Challenges
References
Ensured Continuous Surveillance Despite Sensor Transition Using Control Barrier Functions*-12pt
1 Introduction
2 Problem Formulation
2.1 Control Objectives
3 Selection of ux and uy to Solve Control Objective 1
4 Sufficient Conditions to Solve Control Objective 2
5 Control Laws to Solve Control Objective 2
5.1 Bounds on the Time Derivatives, and a Selection for u( t )
5.2 u ( t ) and u ( t )
6 Simulations
7 Conclusion
References
Nonlinear Stochastic Trajectory Optimization for Centroidal Momentum Motion Generation of Legged Robots*-12pt
1 Introduction
2 Background
2.1 Notation
2.2 Robot Dynamics
2.3 Centroidal Momentum Trajectory Optimization
3 Stochastic Optimal Control for Centroidal Momentum Trajectory Optimization
3.1 Individual Chance Constraints Reformulation
3.2 Deterministic Reformulation of Individual Chance Constraints
3.3 Deterministic Reformulation of SOCP
3.4 SCP with L1 Trust Region Penalty Cost
4 Simulations Results
4.1 Simulations Setup
5 Discussion and Conclusions
References
Reactive Anticipatory Robot Skills with Memory*-12pt
1 Introduction
2 Related Work
3 Background
4 Methods
4.1 Extended System Level Synthesis (eSLS)
4.2 Iterative System Level Synthesis (iSLS)
4.3 Time Correlations Between States
4.4 Adaptation to New Reference States
5 Experiments and Results
5.1 Simulated Task
5.2 Pick-and-Place Task
6 Conclusion
References
ROSE: Robust State Estimation via Online Covariance Adaption
1 Introduction
2 Related Work
3 Problem Formulation
4 Proposed Method
4.1 Method Overview
4.2 Preliminary FGO
4.3 Backward Kalman Smoothing
4.4 Online NCM Estimation
4.5 Main FGO
5 Experiments
5.1 Setup
5.2 Simulation Results
5.3 Hardware Results
6 Conclusions and Future Work
References
What Can Algebraic Topology and Differential Geometry Teach Us About Intrinsic Dynamics and Global Behavior of Robots?*-12pt
1 Introduction
1.1 A Very Short Primer on Robot Dynamics
1.2 Mauptertuis' Principle of Least Action
1.3 Geodesics on Riemannian Manifolds
1.4 A Purely Geometric Perspective on Lagrangian Mechanics
1.5 Algorithms: Shooting Versus Extremizing Arc Length
2 String Relaxation
3 Topological Insights into Periodic Orbits
4 A Case Study: Periodic Orbits of the Double Pendulum
4.1 Toroidal Orbits
4.2 Disk Orbits
4.3 Brake Orbits aka Nonlinear Normal Modes
4.4 Classification and Overview on Periodic Orbits
5 Conclusion
References
Planning
Efficient Task Planning Using Abstract Skills and Dynamic Road Map Matching
1 Introduction
1.1 Background
1.2 Contribution
1.3 Related Work
2 Preliminaries
2.1 Task Planning
2.2 Abstract Skills and Abstraction Keys
3 Practical Demonstration: Grid World Planning Domain
4 Planning with Skills as a Constraint Satisfaction Problem
4.1 Skill Matching
4.2 Action Recovery
4.3 The Joint Problem: Skills as Dynamic Constraints
5 Solving the Problem: A Hierarchical Planning Algorithm
5.1 Abstract-Skill Matching for Reach-Avoid Tasks
5.2 Action Recovery Using Dynamic Task Decomposition
6 Experimental Results
7 Conclusion
References
CALIPSO: A Differentiable Solver for Trajectory Optimization with Conic and Complementarity Constraints*-12pt
1 Introduction
2 Related Work
3 Background
3.1 Trajectory Optimization
3.2 Complementarity Constraints
3.3 Linear Independence Constraint Qualification
3.4 Augmented Lagrangian Methods
3.5 Interior-Point Methods
3.6 Differentiable Optimization
4 CALIPSO
4.1 Search Direction
4.2 Line Search
4.3 Outer Updates
4.4 Solution Derivatives
4.5 Implementation
5 Examples
5.1 Contact-Implicit Trajectory Optimization
5.2 State-Triggered Constraints
5.3 Model Predictive Control Auto-Tuning
6 Conclusion
References
Clutter-Resilient Autonomous Mobile Robot Navigation with Computationally Efficient Free-Space Features*-12pt
1 Introduction
2 Problem Formulation
3 Proposed Solution
3.1 Assumptions and Knowledge
3.2 Designing Motion Tubes
3.3 Sweeping to Detect Collision
3.4 Sampling
3.5 Template Samples to Sensor Space
4 Experimental Validation
5 Conclusions and Discussion
References
Robust-RRT: Probabilistically-Complete Motion Planning for Uncertain Nonlinear Systems
1 Introduction
2 Related Work
3 Problem Formulation
4 Robust-RRT
5 Robust-RRT Is Probabilistically Complete (PC)
6 Insights and Practical Considerations
7 Experiments
8 Conclusion
References
Effort Informed Roadmaps (EIRM*): Efficient Asymptotically Optimal Multiquery Planning by Actively Reusing Validation Effortpg*-12pt
1 Introduction
2 Related Work
3 Effort Informed Roadmaps (EIRM*)
3.1 Notation
3.2 Approximation
3.3 Reverse Search
3.4 Forward Search
4 Experiments and Results
4.1 Abstract Scenarios
4.2 Robotic Scenario
4.3 Initial Solutions
4.4 Construction Scenario
5 Discussion
5.1 Initial Solution Time
5.2 Objective Value
6 Conclusion
References
Author Index