Slide 1
Slide 1 text
A Modern Perspective from “Dual Citizenship” in
Academia and Industry
Avik De
Cofounder & CTO, Ghost Robotics (Philadelphia)
Postdoc (Harvard SEAS); Ph.D. (UPenn 2017)
Slide 2
Slide 2 text
Fielded legged robotics: status update
Vision-60 (35kg, 12dof)
Research Scientific Commercial
Jerboa (3kg, 4dof)
Minitaur (6kg, 8dof direct drive)
Slide 3
Slide 3 text
Perspectives on (legged) robot control
[Raibert (1986)]
Simple controllers →
Complex systems/tasks?
Generalizable solutions
• Across tasks
• Across platforms
[Ting et. al. Neuron (2015)]
In Biology In Robotics
[Seyfarth (2015) [De (2017)]
[Schaal (2006)]
[Posa et al (2014)]
[Boston Dynamics (2019)]
Slide 4
Slide 4 text
Thesis: Modularity in Robot Control (Reduction + Composition)
Classical (analytical) view
Templates are themselves compositions
[Full and Koditschek (1999)] “Templates and anchors…”
Slide 5
Slide 5 text
Applications
[Shamsah, De, Kod (IROS 2018)]
[De, Kod in IJRR (2018)]
Minitaur bounding, trotting, pacing, pronking Jerboa tailed hopping
[De, Kod (ICRA 2014)]
Slide 6
Slide 6 text
Ghost Robotics technical direction: modularity through hierarchy
?
Abstraction benefits
• Reduced
(re)development
• Computational
simplification
• Model-robustness
Spirit 40
Slide 7
Slide 7 text
Consequence on development: GR Vision 60 design history
2018 2019
v1.0 v2.0
v3.0
Q1 Q1
v3.5
Q2 Q2
Q3 Q3
Q4 Q4
Slide 8
Slide 8 text
Reduction (anchoring) exposes templates: reactive quadrupedal walking
Event-driven coupled
swing leg oscillators
Trot walk
Crawl walk
Slide 9
Slide 9 text
Dimension reduction and “reflexes” for perturbation rejection
Slide 10
Slide 10 text
Details of these reflexes make a huge difference in practice: outdoors
“Details”
• Slope estimation
• Slip detection and
handling
• Stubbing detection
and handling
• Early/late contact
handling
• “Re-swing” reflexes
• …
Slide 11
Slide 11 text
Details of these reflexes make a huge difference in practice: indoors
Slide 12
Slide 12 text
Can go further: templates are inevitable
• Subject to anchoring posture control,
• With sufficient actuated DOFs,
• (degrades gracefully with fewer)
• Reduced dynamics at least contain IP.
[De, Topping, Kod (in prep)]
Back from commercial to research!
Slide 13
Slide 13 text
Emphasize scientific fundamentals: push on efficiency
* Estimated from public specs
** MIT cheetah is larger, “actuators only”
mechanical (not total) COT, carrying a small battery
[Seok et. al. (2015)]
Robot Cost of Transport
Vision 60 v3.5 0.80
Vision 60 v4.0 0.54
Boston Dynamics Spot 0.91*
MIT Cheetah 0.46**
Type Power (W)
Motors Mechanical 210
Low level electronics Electrical 20
Blind locomotion Algorithmic <1
Gait planner Algorithmic <5
Autonomy Algorithmic 15
GR Vision 60 v3.5 power budget
Slide 14
Slide 14 text
Perspectives from mixing research with commercial viability
Why do we need Universities?
• Training
• “Five-year problems”
• Intellectual boldness
Why do we need companies?
• Motivation
• Platforms
• $$$
• “Details” reveal research problems
[De (2017)]
BD Spot GR Vision ANYmal
[Boston Dynamics (2019)]
Slide 15
Slide 15 text
Impact on the individual
What does an (academic/researcher/engineer)
of the future look like?
• Beneficial to draw on both
• Unclear
• Incentive structures around IP
• Trade secrets vs. publishing
• Entities like GRASP can help by defining next-generation norms!
Academia Company
Individual