HBP Tea and Slides

Brain-inspired control systems

Jens E. Pedersen

<jeped@kth.se>

Agenda

... how we can train brain-inspired control systems on neuromorphic hardware

... by exploiting the growing knowledge within both neuroscience and computer science

1. The NeuroComputing Systems lab at KTH
2. Algorithms and brains
3. Robots and brains
4. What happens next?

Shallow talk: 10 minutes

Focus: brief overview on how I see the current scene and where to go from there

Hope: insight into the work of the HBP and discussion

1. The NeuroComputing Systems (NCS) lab at KTH

Discovering key principles by which brains work and implementing them in technical systems that intelligently interact with their environment.

Examples

• Analog VLSI Circuits to implement neural processing
• Sensors, such as Silicon Retinae or Cochleae
• Distributed Adaptive Control
• Massively Parallel Self-Growing Computing Architectures
• Neuro-Electronic Implants, Rehabilitation

2. Algorithms and brains

2.1 Algorithms for computers

Source: Wikipedia

$\models$ Algorithms are unambiguous

2.2 Algorithms for the real world

Source: Harry Potter and the Sorcerer's Stone

2.3 Algorithms for organisms

Source: David Rogers, Vanderbilt University

2.4 Algorithm in the brain

• Modern neuroscience pinpoints specific tasks and their (partial) solutions in the brains

DAC-h3: A Proactive Robot Cognitive Architecture to Acquire and Express Knowledge About the World and the Self

3. Soft robotics and neuromorphic computing

3.1 Algorithms for robots

Does one thing and one thing only. Just like algorithms, because they are controlled by algorithms

Moving parts = liability

3D printers good examples because they are diverse

Source: Makerbot, Financial Times

Does one thing and one thing only. Just like algorithms, because they are controlled by algorithms

Moving parts = liability

3D printers good examples because they are diverse

Source: DARPA robotics challenge 2015

What about building robots that can deal with uncertainty?

What systems do we know that deal with uncertainty? The brain

3.2 Brain-inspired controller for the Baxter robot

Source: https://ieeexplore.ieee.org/document/8880621

Source: Baxter

This has been reproduced in other settings, like this experiment from Milano

However, these robots are still hard and unmoveable

3.3 Myorobotic arm

Source: NCS, KTH, https://ieeexplore.ieee.org/document/7553551

Problem: Very difficult to control

Solution: Brain-inspired controls

Source: NCS, KTH, https://ieeexplore.ieee.org/document/7553551

4. What's next? Integrating neuroscience and neurorobotics

• Energy-efficient, scalable algorithms and hardware
  • to implement neuronal algorithms in real-time
  • to interpret (neuronal) sensory input in real-time

Opens for a host of possible hardware options

• Event-based sensors
  • Dynamic vision systems (DVS)
  • Neuro-prosthetics

• Neuronal recordings (EMG / EEG) to create motor signals in real-time

• Robotic interaction within their environment

HBP WP3: Adaptive networks for cognitive architectures: from advanced learning to neurorobotics and neuromorphic applications

WPO3.1: Enhanced real-world task performance through biologically plausible adaptive cognitive architectures running on neuromorphic hardware and closed-loop neuro-robotics platform

Next for us: closed-loop systems interacting and working with humans

• Revolutionary; industrial robots are lethal for humans

HBP Tea and Slides

Brain-inspired control systems

Jens E. Pedersen

<jeped@kth.se>

Credits