报告学习
JaIme (CAS - UTS)
Motivation - Pipes fail (metallic)
- NDT sensing (do not break the pipes)
- 3D laser cameras for GTruth
Sensing - Data Analytics Sensor modeling
Network: Simulations (Training data) + sensor data -> gt (finally prediction)
Robotic design for pipe inspection
Data Modelling - ATP Framework
Probablistic model
Project1: Assistive Robotics: Intelligent wheelchair - no map
Laser-based no map
- CONCEPT: Users drive the wheelchair with the help of assistive robotics: Shared control between the user and wheelchair (real-time)
Project2: Smart Hoist
Project3: Autonomous navigation in rugged terrains
- ZMP-based stable path planning
Continuous Mapping and exploration with Gaussian processes
Project4: Indoor people tracking with RF (BLE)
HRI
Decoupled observation models (Learning mobility aid assistance)
- AIM: provide intelligent assistance to powered mobility devices
- Build model from disabled (experts)
- expert users data collection
- expert modelling: CNN heapmap & GP joystick heatmap model - JOYSTICK as a input not directly control the wheelchair
- Assisting new users
Assistive wheelchair (EXPLORATION)
Expert data collection
- simulated 2D environment
- record odom, LIDAR and joystick inputs
- record safe joystick (collision avoidance)
CNN model (environment)
- Local free-space map -> likelihood
- Training data: free-space
- Goal: end point in local free-space -> likelihood map
User model (GP - Gaussian Processes)
- Add preknowledge more uncertainty in the GP model
- Distance filter
Planning - Qlearning (or any MDP path planning tool)
- Heatmap for Path Planning
- CNN heatmap + User heatmap + obstacle safety map