Resilient Machines Through Continuous Self-Modeling

By Josh C. Bongard, V. Zykov, Hod Lipson. 2006

801 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Biomedical Engineering, Mechanical Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Biomedical Engineering

In Mechanical Engineering

Citations in the corpus

Sim-to-Real: Learning Agile Locomotion For Quadruped Robots (2018, methodology): “ully measured physical parameters were used in simulation. These physical parameters can also be optimized by matching the robot behaviors in the simulated and the real world [34, 35]. Bongard et al. [36] used the actuation-sensation relationship to build and reﬁne a simulation through continuous se”
Learning agile and dynamic motor skills for legged robots (2019, background): “The first is to improve simulation fidelity either analytically or in a data-driven way; the latter is also known as system identification [27–32].”
NoRML: No-Reward Meta Learning (2019, neutral): “Separate from the meta learning literature, many other works have considered the problem of adapting to varying dynamics through,e.g. adaptiveinversecontrol[34],self-modeling[2],Bayesian optimization[6],oronlinesystemidentification[18,35].”
Learning to Predict Without Looking Ahead: World Models Without Forward Prediction (2019, background): “These algorithms are robust and will adapt to constraints such as ours while still solving the given task [7, 33].”
Morphological change in machines accelerates the evolution of robust behavior (2011, background): “This work suggests that the automated design, manufacture (10), and deployment (8) of adaptive machines should allow for neurological and morphological change at ontogenetic and phylogenetic time scales.”
Neuromodulation Improves the Evolution of Forward Models (2016, background): “Being able to predict the consequences of actions could grant robots the ability to better adapt to changes to their environments or bodies; specifically, they could mentally simulate possible movements to select an appropriate one without having to incur the costs and hazards of trying each action ”
Towards an Ethical Robot: Internal Models, Consequences and Ethical Action Selection (2014, background): “[2] describe a 4-legged starfish like robot that makes use of explicit internal simulation, both to enable the robot to learn it’s own body morphology and control, and notably allow the robot to recover from physical damage by learning the new morphology following the damage.”
Fast Model Identification via Physics Engines for Data-Efficient Policy Search (2017, positive): “Optimizing the simulated model parameters to match actual robot data can also help model robot damage [Bongard et al. , 2006].”
Online Extreme Evolutionary Learning Machines (2014, neutral): “This should allow for a robot to easily adapt to damage (Bongard et al., 2006) or to changes in mor-”
The Transferability Approach: Crossing the Reality Gap in Evolutionary Robotics (2013, methodology): “It is also argued in [7] that accurate quantitative comparisons between two sensor time series is difficult because small initial disparities can quickly lead to very different signals, which can lead to prefer external measures for physical robots [8].”
Gait optimization for roombots modular robots — Matching simulation and reality (2013, background): “The parameter space exploration requires experiments running in real time and unless many robots with wellmatched behavior are available the optimization process cannot be parallelized.”
Reset-free Trial-and-Error Learning for Robot Damage Recovery (2016, methodology): “Classic approaches to fault tolerance rely on updating the model of the robot, either directly with selfdiagnosis [37], or indirectly with machine learning [38, 5]; the model is then used for planning and/or control.”