Neural Modular Control for Embodied Question Answering

arXiv.org Artificial Intelligence

We present a modular approach for learning policies for navigation over long planning horizons from language input. Our hierarchical policy operates at multiple timescales, where the higher-level master policy proposes subgoals to be executed by specialized sub-policies. Our choice of subgoals is compositional and semantic, i.e. they can be sequentially combined in arbitrary orderings, and assume human-interpretable descriptions (e.g. 'exit room', 'find kitchen', 'find refrigerator', etc.). We use imitation learning to warm-start policies at each level of the hierarchy, dramatically increasing sample efficiency, followed by reinforcement learning. Independent reinforcement learning at each level of hierarchy enables sub-policies to adapt to consequences of their actions and recover from errors. Subsequent joint hierarchical training enables the master policy to adapt to the sub-policies.


Visual Semantic Navigation using Scene Priors

arXiv.org Artificial Intelligence

Figure 1: Our goal is to use scene priors to improve navigation in unseen scenes and towards novel objects. The most likely locations are shown with the orange box. How do humans navigate to target objects in novel scenes? Do we use the semantic/functional priors we have built over years to efficiently search and navigate? For example, to search for mugs, we search cabinets near the coffee machine and for fruits we try the fridge. In this work, we focus on incorporating semantic priors in the task of semantic navigation. We propose to use Graph Convolutional Networks for incorporating the prior knowledge into a deep reinforcement learning framework. The agent uses the features from the knowledge graph to predict the actions. For evaluation, we use the AI2-THOR framework. Our experiments show how semantic knowledge improves performance significantly. More importantly, we show improvement in generalization to unseen scenes and/or objects.


DIViS: Domain Invariant Visual Servoing for Collision-Free Goal Reaching

arXiv.org Artificial Intelligence

Robots should understand both semantics and physics to be functional in the real world. While robot platforms provide means for interacting with the physical world they cannot autonomously acquire object-level semantics without needing human. In this paper, we investigate how to minimize human effort and intervention to teach robots perform real world tasks that incorporate semantics. We study this question in the context of visual servoing of mobile robots and propose DIViS, a Domain Invariant policy learning approach for collision free Visual Servoing. DIViS incorporates high level semantics from previously collected static human-labeled datasets and learns collision free servoing entirely in simulation and without any real robot data. However, DIViS can directly be deployed on a real robot and is capable of servoing to the user-specified object categories while avoiding collisions in the real world. DIViS is not constrained to be queried by the final view of goal but rather is robust to servo to image goals taken from initial robot view with high occlusions without this impairing its ability to maintain a collision free path. We show the generalization capability of DIViS on real mobile robots in more than 90 real world test scenarios with various unseen object goals in unstructured environments. DIViS is compared to prior approaches via real world experiments and rigorous tests in simulation. For supplementary videos, see: \href{https://fsadeghi.github.io/DIViS}{https://fsadeghi.github.io/DIViS}


Search on the Replay Buffer: Bridging Planning and Reinforcement Learning

arXiv.org Artificial Intelligence

The history of learning for control has been an exciting back and forth between two broad classes of algorithms: planning and reinforcement learning. Planning algorithms effectively reason over long horizons, but assume access to a local policy and distance metric over collision-free paths. Reinforcement learning excels at learning policies and the relative values of states, but fails to plan over long horizons. Despite the successes of each method in various domains, tasks that require reasoning over long horizons with limited feedback and high-dimensional observations remain exceedingly challenging for both planning and reinforcement learning algorithms. Frustratingly, these sorts of tasks are potentially the most useful, as they are simple to design (a human only need to provide an example goal state) and avoid reward shaping, which can bias the agent towards finding a sub-optimal solution. We introduce a general control algorithm that combines the strengths of planning and reinforcement learning to effectively solve these tasks. Our aim is to decompose the task of reaching a distant goal state into a sequence of easier tasks, each of which corresponds to reaching a subgoal. Planning algorithms can automatically find these waypoints, but only if provided with suitable abstractions of the environment -- namely, a graph consisting of nodes and edges. Our main insight is that this graph can be constructed via reinforcement learning, where a goal-conditioned value function provides edge weights, and nodes are taken to be previously seen observations in a replay buffer. Using graph search over our replay buffer, we can automatically generate this sequence of subgoals, even in image-based environments. Our algorithm, search on the replay buffer (SoRB), enables agents to solve sparse reward tasks over one hundred steps, and generalizes substantially better than standard RL algorithms.


Learning Exploration Policies for Navigation

arXiv.org Artificial Intelligence

Numerous past works have tackled the problem of task-driven navigation. But, how to effectively explore a new environment to enable a variety of downstream tasks has received much less attention. In this work, we study how agents can autonomously explore realistic and complex 3D environments without the context of task-rewards. We propose a learning-based approach and investigate different policy architectures, reward functions, and training paradigms. We find that use of policies with spatial memory that are bootstrapped with imitation learning and finally finetuned with coverage rewards derived purely from on-board sensors can be effective at exploring novel environments. We show that our learned exploration policies can explore better than classical approaches based on geometry alone and generic learning-based exploration techniques. Finally, we also show how such task-agnostic exploration can be used for downstream tasks. Imagine your first day at a new workplace. If you are like most people, the first task you set for yourself is to become familiar with the office so that the next day when you have to attend meetings and perform tasks, you can navigate efficiently and seamlessly. To achieve that goal, you explore your office without the task context of target locations you have to reach and build a generic understanding of space. This step of task-independent exploration is quite critical yet often ignored in current approaches for navigation. When it comes to navigation, currently there are two paradigms: (a) geometric reconstruction and path-planning based approaches (Hartley & Zisserman, 2003; Thrun et al., 2005; LaValle, 2006), and (b) learning-based approaches (Mirowski et al., 2017; Gupta et al., 2017; Savinov et al., 2018; Zhu et al., 2017).