Knowledge Vault 4 /98 - AI For Good 2024
Robotic autonomy in the wild
Kostas Alexis
< Resume Image >
Link to IA4Good VideoView Youtube Video

Concept Graph & Resume using Claude 3 Opus | Chat GPT4o | Llama 3:

graph LR classDef robots fill:#d4d4f9, font-weight:bold, font-size:14px classDef sensors fill:#f9d4d4, font-weight:bold, font-size:14px classDef resilience fill:#d4f9d4, font-weight:bold, font-size:14px classDef applications fill:#f9f9d4, font-weight:bold, font-size:14px classDef funding fill:#f9d4f9, font-weight:bold, font-size:14px A[Robotic autonomy in
the wild] --> B[Robots explore
extreme environments,
inspect infrastructure. 1] A --> C[Legged robot,
drone explore
stage area. 2] A --> D[Robots use sensors
like lasers,
radars, vision. 3] A --> E[Robots 3D model
shows compartment,
room. 4] A --> F[Safety pilot
present, intervenes
if needed. 5] A --> G[Robots lack
prior environmental
information. 6] B --> H[Soft drones
sustain collisions,
fit narrow spaces. 15] B --> I[Quadruped explored
mine, mapped
kilometers. 16] B --> J[Robots generate
3D models
autonomously. 17] D --> K[Sensing: robot
perceives with
sensors. 10] K --> L[Data fusion:
combining sensor
data effectively. 11] L --> M[Planning:
determining robots
safe actions. 12] M --> N[Actuation:
executing planned
robot motions. 13] G --> O[Moravecs paradox:
human tasks hard
for AI. 7] O --> P[Humans generalize,
understand risk,
stay resourceful. 8] P --> Q[Goal: resilience
across robot
technology stack. 9] Q --> R[Resilience key:
handling disturbances,
uncertainty. 28] B --> S[Physical embodiment
crucial, not just
intelligence. 14] S --> T[Specializing robot design
for environments
promising. 21] T --> U[Co-designing robot
body, brain inspired
by nature. 22] U --> V[Resilient autonomy:
challenging, exciting
research. 23] I --> W[Applications: caves,
forests, undersea. 18] W --> X[Inspecting mines,
ballast tanks,
nuclear plants. 19] W --> Y[Practical robots
inspect infrastructure,
suppress wildfires. 24] Y --> Z[Autonomous missions in
natural, industrial
settings. 30] A --> AA[Progress made,
robots not seamless
yet. 20] AA --> BB[Robots shipped
from Norway
for demo. 26] BB --> CC[Drone built in lab,
carried by
quadruped. 27] AA --> DD[Public funding
supports fundamental
research. 25] D --> EE[Fusing multiple
sensor data
robustly. 29] class A,BB,CC,DD funding class B,H,I,J,S,T,U,V resilience class C sensors class D,K,L,M,N,EE robots class F,O,P,Q,R applications class G,W,X,Y,Z applications

Resume:

1.- Autonomous robots can explore extreme environments like caves and glaciers, and inspect critical infrastructure.

2.- Demo: Legged robot and drone working together to explore a two-compartment area on stage.

3.- Robots rely on onboard sensors like laser scanners, radars, accelerometers, gyroscopes and vision systems.

4.- 3D model reconstructed by robot shows the small compartment and overall room.

5.- Safety pilot present but not actively intervening unless needed.

6.- Robots not given prior information about the environment.

7.- Moravec's paradox: What's hard for humans is easy for AI/robots and vice versa.

8.- Humans generalize well, understand risk, remain resourceful despite uncertainty and disturbances.

9.- Goal is to instill resilience across the robot technology stack.

10.- Sensing: How robot perceives environment with various sensors.

11.- Data fusion: Combining sensor data to understand and reason about the world.

12.- Planning: Determining safe, useful actions for the robot to take.

13.- Actuation: Executing planned motions and delivering actions.

14.- Physical embodiment is crucial in robotics, not just intelligence.

15.- Soft drones can sustain collisions and fit through narrow openings.

16.- Quadruped autonomously explored underground mine and mapped it over several kilometers.

17.- Robots can enter unknown areas and generate detailed 3D models without human teleoperation.

18.- Applications in extreme natural environments like caves, forests, under the sea.

19.- Industrial applications inspecting mines, ship ballast tanks, nuclear plants.

20.- Progress made but robots not quite able to seamlessly operate in the wild yet.

21.- Specializing robot design to specific environments and niches is promising.

22.- Computational co-design of robot body and brain, taking inspiration from nature.

23.- Resilient autonomy is challenging but an exciting research endeavor.

24.- Near-term practical robots can inspect infrastructure or even help suppress wildfires.

25.- Public sector funding is crucial to support fundamental research before excellent results emerge.

26.- Speaker's team shipped robots from Norway to Switzerland for the demo.

27.- Drone was built in their lab, modified ANYbotics quadruped carried it.

28.- Resilience is key - dealing with disturbances, uncertainty, generalizing from experience.

29.- Fusing data from multiple sensor modalities to robustly understand the world.

30.- Enabling useful autonomous missions in natural and industrial settings without human teleoperation.

Knowledge Vault built byDavid Vivancos 2024