Learning Continuous Control Policies for Information-Theoretic Active Perception

This paper proposes a method for learning continuous control policies for exploration and active landmark localization. We consider a mobile robot detecting landmarks within a limited sensing range, and tackle the problem of learning a control policy that maximizes the mutual information between the landmark states and the sensor observations. We employ a Kalman filter to convert the partially observable problem in the landmark states to a Markov decision process (MDP), a differentiable field of view to shape the reward function, and an attention-based neural network to represent the control policy. The approach is combined with active volumetric mapping to promote environment exploration in addition to landmark localization. The performance is demonstrated in several simulated landmark localization tasks in comparison with benchmark methods.

Authors

• Pengzhi Yang
• Yuhan Liu
• Shumon Koga
• Arash Asgharivaskasi
• Nikolay Atanasov 0001

Keywords

• Location awareness
• Shape
• Neural networks
• Process control
• Markov processes
• Robot sensing systems
• Sensors
• Neural Network
• Field Of View
• Mutual Information
• Kalman Filter
• Reward Function
• Markov Decision Process
• Landmark Localization
• Exploration Of Environment
• Attention-based Neural Network
• Hyperparameters
• Optimal Control
• Control Input
• Mean Absolute Error
• Fisher Information
• Linear Velocity
• Policy Learning
• Sensor Model
• Sensor Noise
• Observation Space
• Simultaneous Localization And Mapping
• Proximal Policy Optimization
• Limited Field Of View
• Stochastic Control
• Robot State
• Conditional Mutual Information
• Deep Q-learning
• End Of Episode
• Policy Gradient Method
• Imitation Learning
• Sufficient Statistics