Offline reinforcement learning (RL) allows agents to learn effective, return-maximizing policies from a static dataset. Three popular algorithms for offline RL are Conservative Q-Learning (CQL), Behavior Cloning (BC), and Decision Transformer (DT), from the class of Q-Learning, Imitation Learning, and Sequence Modeling respectively. A key open question is: which algorithm is preferred under what conditions? We study this question empirically by exploring the performance of these algorithms across the commonly used D4RL and Robomimic benchmarks. We design targeted experiments to understand their behavior concerning data suboptimality, task complexity, and stochasticity. Our key findings are: (1) DT requires more data than CQL to learn competitive policies but is more robust; (2) DT is a substantially better choice than both CQL and BC in sparse-reward and low-quality data settings; (3) DT and BC are preferable as task horizon increases, or when data is obtained from human demonstrators; and (4) CQL excels in situations characterized by the combination of high stochasticity and low data quality. We also investigate architectural choices and scaling trends for DT on \textsc{atari} and D4RL and make design/scaling recommendations. We find that scaling the amount of data for DT by 5x gives a 2.5x average score improvement on Atari.

Offline Goal-Conditioned Reinforcement Learning (GCRL) is tasked with learning to achieve multiple goals in an environment purely from offline datasets using sparse reward functions. Offline GCRL is pivotal for developing generalist agents capable of leveraging pre-existing datasets to learn diverse and reusable skills without hand-engineering reward functions. However, contemporary approaches to GCRL based on supervised learning and contrastive learning are often suboptimal in the offline setting. An alternative perspective on GCRL optimizes for occupancy matching, but necessitates learning a discriminator, which subsequently serves as a pseudo-reward for downstream RL. Inaccuracies in the learned discriminator can cascade, negatively influencing the resulting policy. We present a novel approach to GCRL under a new lens of mixture-distribution matching, leading to our discriminator-free method: SMORe. The key insight is combining the occupancy matching perspective of GCRL with a convex dual formulation to derive a learning objective that can better leverage suboptimal offline data. SMORe learns *scores* or unnormalized densities representing the importance of taking an action at a state for reaching a particular goal. SMORe is principled and our extensive experiments on the fully offline GCRL benchmark composed of robot manipulation and locomotion tasks, including high-dimensional observations, show that SMORe can outperform state-of-the-art baselines by a significant margin.

Model-based reinforcement learning (RL) has shown great promise due to its sample efficiency, but still struggles with long-horizon sparse-reward tasks, especially in offline settings where the agent learns from a fixed dataset. We hypothesize that model-based RL agents struggle in these environments due to a lack of long-term planning capabilities, and that planning in a temporally abstract model of the environment can alleviate this issue. In this paper, we make two key contributions: 1) we introduce an offline model-based RL algorithm, IQL-TD-MPC, that extends the state- of-the-art Temporal Difference Learning for Model Predictive Control (TD-MPC) with Implicit Q-Learning (IQL); and 2) we propose to use IQL-TD-MPC as a Manager in a hierarchical setting with any off-the-shelf offline RL algorithm as a Worker. More specifically, we pre-train a temporally abstract IQL-TD-MPC Manager to predict "intent embeddings", which roughly correspond to subgoals, via planning. We show that augmenting state representations with intent embeddings generated by an IQL-TD-MPC manager significantly improves off-the-shelf offline RL agents' performance on some of the most challenging D4RL benchmark tasks. For instance, the offline RL algorithms AWAC, TD3-BC, DT, and CQL all get zero or near-zero normalized evaluation scores on the medium and large antmaze tasks, while our modification gives an average score over 40.

Efficient planning in continuous state and action spaces is fundamentally hard, even when the transition model is deterministic and known. One way to alleviate this challenge is to perform bilevel planning with abstractions, where a high-level search for abstract plans is used to guide planning in the original transition space. Previous work has shown that when state abstractions in the form of symbolic predicates are hand-designed, operators and samplers for bilevel planning can be learned from demonstrations. In this work, we propose an algorithm for learning predicates from demonstrations, eliminating the need for manually specified state abstractions. Our key idea is to learn predicates by optimizing a surrogate objective that is tractable but faithful to our real efficient-planning objective. We use this surrogate objective in a hill-climbing search over predicate sets drawn from a grammar. Experimentally, we show across four robotic planning environments that our learned abstractions are able to quickly solve held-out tasks, outperforming six baselines.

In realistic applications of object search, robots will need to locate target objects in complex environments while coping with unreliable sensors, especially for small or hard-to-detect objects. In such settings, correlational information can be valuable for planning efficiently. Previous approaches that consider correlational information typically resort to ad-hoc, greedy search strategies. We introduce the Correlational Object Search POMDP (COS-POMDP), which models correlations while preserving optimal solutions with a reduced state space. We propose a hierarchical planning algorithm to scale up COS-POMDPs for practical domains. Our evaluation, conducted with the AI2-THOR household simulator and the YOLOv5 object detector, shows that our method finds objects more successfully and efficiently compared to baselines, particularly for hard-to-detect objects such as srub brush and remote control.

In robotic domains, learning and planning are complicated by continuous state spaces, continuous action spaces, and long task horizons. In this work, we address these challenges with Neuro-Symbolic Relational Transition Models (NSRTs), a novel class of models that are data-efficient to learn, compatible with powerful robotic planning methods, and generalizable over objects. NSRTs have both symbolic and neural components, enabling a bilevel planning scheme where symbolic AI planning in an outer loop guides continuous planning with neural models in an inner loop. Experiments in four robotic planning domains show that NSRTs can be learned very data-efficiently, and then used for fast planning in new tasks that require up to 60 actions and involve many more objects than were seen during training.

We address the problem of efficient exploration for transition model learning in the relational model-based reinforcement learning setting without extrinsic goals or rewards. Inspired by human curiosity, we propose goal-literal babbling (GLIB), a simple and general method for exploration in such problems. GLIB samples relational conjunctive goals that can be understood as specific, targeted effects that the agent would like to achieve in the world, and plans to achieve these goals using the transition model being learned. We provide theoretical guarantees showing that exploration with GLIB will converge almost surely to the ground truth model. Experimentally, we find GLIB to strongly outperform existing methods in both prediction and planning on a range of tasks, encompassing standard PDDL and PPDDL planning benchmarks and a robotic manipulation task implemented in the PyBullet physics simulator. Video: https://youtu.be/F6lmrPT6TOY Code: https://git.io/JIsTB

Real-world planning problems often involve hundreds or even thousands of objects, straining the limits of modern planners. In this work, we address this challenge by learning to predict a small set of objects that, taken together, would be sufficient for finding a plan. We propose a graph neural network architecture for predicting object importance in a single inference pass, thus incurring little overhead while greatly reducing the number of objects that must be considered by the planner. Our approach treats the planner and transition model as black boxes, and can be used with any off-the-shelf planner. Empirically, across classical planning, probabilistic planning, and robotic task and motion planning, we find that our method results in planning that is significantly faster than several baselines, including other partial grounding strategies and lifted planners. We conclude that learning to predict a sufficient set of objects for a planning problem is a simple, powerful, and general mechanism for planning in large instances. Video: https://youtu.be/FWsVJc2fvCE Code: https://git.io/JIsqX

Robotic planning problems in hybrid state and action spaces can be solved by integrated task and motion planners (TAMP) that handle the complex interaction between motion-level decisions and task-level plan feasibility. TAMP approaches rely on domain-specific symbolic operators to guide the task-level search, making planning efficient. In this work, we formalize and study the problem of operator learning for TAMP. Central to this study is the view that operators define a lossy abstraction of the transition model of a domain. We then propose a bottom-up relational learning method for operator learning and show how the learned operators can be used for planning in a TAMP system. Experimentally, we provide results in three domains, including long-horizon robotic planning tasks. We find our approach to substantially outperform several baselines, including three graph neural network-based model-free approaches from the recent literature. Video: https://youtu.be/iVfpX9BpBRo. Code: https://git.io/JCT0g

We address the problem of effectively composing skills to solve sparse-reward tasks in the real world. Given a set of parameterized skills (such as exerting a force or doing a top grasp at a location), our goal is to learn policies that invoke these skills to efficiently solve such tasks. Our insight is that for many tasks, the learning process can be decomposed into learning a state-independent task schema (a sequence of skills to execute) and a policy to choose the parameterizations of the skills in a state-dependent manner. For such tasks, we show that explicitly modeling the schema's state-independence can yield significant improvements in sample efficiency for model-free reinforcement learning algorithms. Furthermore, these schemas can be transferred to solve related tasks, by simply re-learning the parameterizations with which the skills are invoked. We find that doing so enables learning to solve sparse-reward tasks on real-world robotic systems very efficiently. We validate our approach experimentally over a suite of robotic bimanual manipulation tasks, both in simulation and on real hardware. See videos at http://tinyurl.com/chitnis-schema.

Multi-object manipulation problems in continuous state and action spaces can be solved by planners that search over sampled values for the continuous parameters of operators. The efficiency of these planners depends critically on the effectiveness of the samplers used, but effective sampling in turn depends on details of the robot, environment, and task. Our strategy is to learn functions called speciatizers that generate values for continuous operator parameters, given a state description and values for the discrete parameters. Rather than trying to learn a single specializer for each operator from large amounts of data on a single task, we take a modular meta-learning approach. We train on multiple tasks and learn a variety of specializers that, on a new task, can be quickly adapted using relatively little data - thus, our system learns quickly to plan quickly using these specializers. We validate our approach experimentally in simulated 3D pick-and-place tasks with continuous state and action spaces. Visit http://tinyurl.com/chitnis-icra-19 for a supplementary video.

In many applications that involve processing high-dimensional data, it is important to identify a small set of entities that account for a significant fraction of detections. Rather than formalize this as a clustering problem, in which all detections must be grouped into hard or soft categories, we formalize it as an instance of the frequent items or heavy hitters problem, which finds groups of tightly clustered objects that have a high density in the feature space. We show that the heavy hitters formulation generates solutions that are more accurate and effective than the clustering formulation. In addition, we present a novel online algorithm for heavy hitters, called HAC, which addresses problems in continuous space, and demonstrate its effectiveness on real video and household domains.

In partially observed environments, it can be useful for a human to provide the robot with declarative information that represents probabilistic relational constraints on properties of objects in the world, augmenting the robot's sensory observations. For instance, a robot tasked with a search-and-rescue mission may be informed by the human that two victims are probably in the same room. An important question arises: how should we represent the robot's internal knowledge so that this information is correctly processed and combined with raw sensory information? In this paper, we provide an efficient belief state representation that dynamically selects an appropriate factoring, combining aspects of the belief when they are correlated through information and separating them when they are not. This strategy works in open domains, in which the set of possible objects is not known in advance, and provides significant improvements in inference time over a static factoring, leading to more efficient planning for complex partially observed tasks. We validate our approach experimentally in two open-domain planning problems: a 2D discrete gridworld task and a 3D continuous cooking task. A supplementary video can be found at http://tinyurl.com/chitnis-iros-18.

Tasks in mobile manipulation planning often require thousands of individual motions to complete. Such tasks require reasoning about complex goals as well as the feasibility of movements in configuration space. In discrete representations, planning complexity is exponential in the length of the plan. In mobile manipulation, parameters for an action often draw from a continuous space, so we must also cope with an infinite branching factor. Task and motion planning (TAMP) methods integrate logical search over high-level actions with geometric reasoning to address this challenge. We present an algorithm that searches the space of possible task and motion plans and uses statistical machine learning to guide the search process. Our contributions are as follows: 1) we present a complete algorithm for TAMP; 2) we present a randomized local search algorithm for plan refinement that is easily formulated as a Markov decision process (MDP); 3) we apply reinforcement learning (RL) to learn a policy for this MDP; 4) we learn from expert demonstrations to efficiently search the space of high-level task plans, given options that address different (potential) infeasibilities; and 5) we run experiments to evaluate our system in a variety of simulated domains. We show significant improvements in performance over prior work.

We consider the problem of refining an abstract task plan into a motion trajectory. Task and motion planning is a hard problem that is essential to long-horizon mobile manipulation. Many approaches divide the problem into two steps: a search for a task plan and task plan refinement to find a feasible trajectory. We apply sequential quadratic programming to jointly optimize over the parameters in a task plan (e.g., trajectories, grasps, put down locations). We provide two modifications that make our formulation more suitable to task and motion planning. We show how to use movement primitives to reuse previous solutions (and so save optimization effort) without trapping the algorithm in a poor basin of attraction. We also derive an early convergence criterion that lets us quickly detect unsatisfiable constraints so we can re-initialize their variables. We present experiments in a navigation amongst movable objects domain and show substantial improvement in cost over a backtracking refinement algorithm.

Recent work in neural machine translation has shown promising performance, but the most effective architectures do not scale naturally to large vocabulary sizes.We propose and compare three variable-length encoding schemes that represent a large vocabulary corpus using a much smaller vocabulary with no loss in information.Common words are unaffected by our encoding, but rare words are encoded using a sequence of two pseudo-words.Our method is simple and effective: it requires no complete dictionaries, learning procedures, increased training time, changes to the model, or new parameters.Compared to a baseline that replaces all rare words with an unknown word symbol, our best variable-length encoding strategy improves WMT English-French translation performance by up to 1.7 BLEU.

The execution of long-horizon tasks under uncertainty is a fundamental challenge in robotics. Recent approaches have made headway on these tasks with an integration of task and motion planning. In this paper, we present Interfaced Belief Space Planning (IBSP): a modular approach to task and motion planning in belief space. We use a task-independent interface layer to combine an off-the-shelf classical planner with motion planning and inference. We determinize the problem under the maximum likelihood observation assumption to obtain a deterministic representation where successful plans generate goal-directed observations. We leverage properties of maximum likelihood observation determinizations to obtain a simple representation of (optimistic) belief space dynamics that is well-suited to planning. Our interface is implemented with standard belief state queries, requiring only the ability to sample, compute unnormalized likelihoods, and compute maximum likelihood states. Our contribution is a novel algorithm for task and motion planning in belief space that has minimal dependence on the details of the inference engine used. IBSP can work with a broad class of black box state estimators, with zero changes to the algorithm. We validate our approach in simulated tasks for the PR2 that account for continuous state, different types of initial state distributions, and negative observations.

The need for combined task and motion planning in robotics is well understood. Solutions to this problem have typically relied on special purpose, integrated implementations of task planning and motion planning algorithms. We propose a new approach that uses off-the-shelf task planners and motion planners and makes no assumptions about their implementation. Doing so enables our approach to directly build on, and benefit from, the vast literature and latest advances in task planning and motion planning. It uses a novel representational abstraction and requires only that failures in computing a motion plan for a high-level action be identifiable and expressible in the form of logical predicates at the task level. We evaluate the approach and illustrate its robustness through a number of experiments using a state-of-the-art robotics simulator and a PR2 robot. These experiments show the system accomplishing a diverse set of challenging tasks such as taking advantage of a tray when laying out a table for dinner and picking objects from cluttered environments where other objects need to be re-arranged before the target object can be reached.