• Corpus ID: 221836419

Adaptive Meta-Learning for Identification of Rover-Terrain Dynamics

@article{Banerjee2020AdaptiveMF,
  title={Adaptive Meta-Learning for Identification of Rover-Terrain Dynamics},
  author={S. Banerjee and James Harrison and P. Michael Furlong and Marco Pavone},
  journal={ArXiv},
  year={2020},
  volume={abs/2009.10191}
}
Rovers require knowledge of terrain to plan trajectories that maximize safety and efficiency. Terrain type classification relies on input from human operators or machine learning-based image classification algorithms. However, high level terrain classification is typically not sufficient to prevent incidents such as rovers becoming unexpectedly stuck in a sand trap; in these situations, online rover-terrain interaction data can be leveraged to accurately predict future dynamics and prevent… 

Figures from this paper

Safe Active Dynamics Learning and Control: A Sequential Exploration–Exploitation Framework
TLDR
This work proposes a practical and theoretically-justified approach to maintaining safety in the presence of dynamics uncertainty that leverages Bayesian meta-learning with last-layer adaptation and motivates two regularizers of last- layer meta- learning models that improve online adaptation capabilities as well as performance by reducing the size of the confidence sets.
SegVisRL: development of a robot's neural visuomotor and planning system for lunar exploration
TLDR
This method combines exteroceptive data from the camera and the goal location information along with proprioceptive information, such as the robot”s angular rotation, and the RL agent is able to directly control the actuator outputs of the rover.
Optimisation of Structured Neural Controller Based on Continuous-Time Policy Gradient
TLDR
This study presents a policy optimisation framework for structured nonlinear control of continuoustime (deterministic) dynamic systems that enables combining the stability, robustness, and physical interpretability of an analytically-derived structure for the feedback controller with the representational flexibility and optimised resulting performance provided by machine learning techniques.
Fractional Binding in Vector Symbolic Representations for Efficient Mutual Information Exploration
TLDR
This work presents an efficient implementation of MI-driven exploration by combining vector symbolic architectures with Bayesian Linear Regression, demonstrating equivalent regret performance to a GP-based approach with memory and time complexity that is constant in the number of samples collected, enabling long-term exploration.

References

SHOWING 1-10 OF 37 REFERENCES
Learning to predict slip for ground robots
TLDR
A generic nonlinear regression framework is proposed in which the terrain type is determined from appearance and then a nonlinear model of slip is learned for a particular terrain type.
Vibration-based terrain classification for planetary exploration rovers
TLDR
A method to classify terrain based on vibrations induced in the rover structure by wheel-terrain interaction during driving, which is robust to lighting variations and has been experimentally validated on a laboratory testbed and on a four-wheeled rover in outdoor conditions.
Terrain parameter estimation and traversability assessment for mobile robots
The estimation of terrain characteristics is an important missions of Martian exploration rovers. Since only limited resources and human supervision are available, efficient and autonomous method of
SPOC: Deep Learning-based Terrain Classification for Mars Rover Missions
TLDR
SPOC has a promising potential for a wider range of future applications, such as the automated discovery of scientifically important terrain features on existing Mars orbital imagery, as well as traversability analysis for future surface missions to small bodies and icy worlds.
Continuous Vehicle Slip Model Identification on Changing Terrains
TLDR
Integrated Perturbative Dynamics is used to successively identify systematic and stochastic models of vehicle slip, a real-time algorithm which works over arbitrary trajectories, with fast convergence along terrain boundaries, which allows for reliable operation of ground vehicles over extended periods of time in changing environments.
MOBILE ROBOT ROUGH-TERRAIN CONTROL (RTC) FOR PLANETARY EXPLORATION
TLDR
A rough-terrain control (RTC) methodology is presented that exploits the actuator redundancy found in multi-wheeled mobile robot systems to improve ground traction and reduce power consumption.
Aiding off-road inertial navigation with high performance models of wheel slip
TLDR
A unique approach of a predictive vehicle slip model in a delayed state extended Kalman filter is presented and it is concluded that the position-denied performance of the compensated system is far superior.
Control Adaptation via Meta-Learning Dynamics
TLDR
This paper presents CAMeLiD, a data-driven approach to learn controllers that can quickly adapt to previously unseen dynamics through a high-capacity neural network trained in a meta-learning manner to infer the context of the dynamics and maintain approximate posteriors over the dynamics model.
Slippage and immobilization detection for planetary exploration rovers via machine learning and proprioceptive sensing
TLDR
This paper addresses the most convenient placement of IMU sensors on the rover chassis such that slippage detection is maximized and demonstrates a proper balance between the benefits of supervised learning algorithms and the advantages of unsupervised learning methods.
Learning to Adapt in Dynamic, Real-World Environments through Meta-Reinforcement Learning
TLDR
This work uses meta-learning to train a dynamics model prior such that, when combined with recent data, this prior can be rapidly adapted to the local context and demonstrates the importance of incorporating online adaptation into autonomous agents that operate in the real world.
...
...