Prof. talk: Mobile Robot

Headline

Probabilitic SKAM formuation
- research: decide (map representation) → model (likelihood) → choose ML/MAP problem → find
- state: velocity (phy), sensors,
  - Map representation
    - depth map, truncated sidgned disance function,
Multi-sensor fusion
- visual only vs visual inertial
- nonlin. least sq.
  - cost: reprojections errors, diff 2D keypoints and 3D projection
- OKVIS, open source
Semantic SLAM
- Semantic fusion
- Input → CNN
- semantic texture for dense tracking
Learning for SLAM
- codeSLAM
Training data set
Future: autnoomous drone with intuitive user interface

Lehr probe
Ebene von robot intelligent
- Moravec's paradox
- wissen & planning: enable introspection, shared representation of map, knowledge
- gap problem: limited robustness
  - deep learning
  - verhalten, regelung durch contruction
Regelungsteknik SISO
- SISO, MIMO
- proportional intergral differenzial regler
  - time diffrentiated measurement
Anticipation & simulation
- PID regler extension
  - reference filter
  - anti-reset-windup
  - pre-control