Nicholas Roy

Allison C. Tam, Neil C. Rabinowitz, Andrew K. Lampinen et al. · deepmind, stanford

Effective exploration is a challenge in reinforcement learning (RL). Novelty-based exploration methods can suffer in high-dimensional state spaces, such as continuous partially-observable 3D environments. We address this challenge by defining novelty using semantically meaningful state abstractions, which can be found in learned representations shaped by natural language. In particular, we evaluate vision-language representations, pretrained on natural image captioning datasets. We show that these pretrained representations drive meaningful, task-relevant exploration and improve performance on 3D simulated environments. We also characterize why and how language provides useful abstractions for exploration by considering the impacts of using representations from a pretrained model, a language oracle, and several ablations. We demonstrate the benefits of our approach in two very different task domains -- one that stresses the identification and manipulation of everyday objects, and one that requires navigational exploration in an expansive world. Our results suggest that using language-shaped representations could improve exploration for various algorithms and agents in challenging environments.

Yongchao Chen, Jacob Arkin, Charles Dawson et al.

For effective human-robot interaction, robots need to understand, plan, and execute complex, long-horizon tasks described by natural language. Recent advances in large language models (LLMs) have shown promise for translating natural language into robot action sequences for complex tasks. However, existing approaches either translate the natural language directly into robot trajectories or factor the inference process by decomposing language into task sub-goals and relying on a motion planner to execute each sub-goal. When complex environmental and temporal constraints are involved, inference over planning tasks must be performed jointly with motion plans using traditional task-and-motion planning (TAMP) algorithms, making factorization into subgoals untenable. Rather than using LLMs to directly plan task sub-goals, we instead perform few-shot translation from natural language task descriptions to an intermediate task representation that can then be consumed by a TAMP algorithm to jointly solve the task and motion plan. To improve translation, we automatically detect and correct both syntactic and semantic errors via autoregressive re-prompting, resulting in significant improvements in task completion. We show that our approach outperforms several methods using LLMs as planners in complex task domains. See our project website https://yongchao98.github.io/MIT-REALM-AutoTAMP/ for prompts, videos, and code.

Yongchao Chen, Jacob Arkin, Yilun Hao et al.

Prompt optimization aims to find the best prompt to a large language model (LLM) for a given task. LLMs have been successfully used to help find and improve prompt candidates for single-step tasks. However, realistic tasks for agents are multi-step and introduce new challenges: (1) Prompt content is likely to be more extensive and complex, making it more difficult for LLMs to analyze errors, (2) the impact of an individual step is difficult to evaluate, and (3) different people may have varied preferences about task execution. While humans struggle to optimize prompts, they are good at providing feedback about LLM outputs; we therefore introduce a new LLM-driven discrete prompt optimization framework PRompt Optimization in Multi-Step Tasks (PROMST) that incorporates human-designed feedback rules to automatically offer direct suggestions for improvement. We also use an extra learned heuristic model that predicts prompt performance to efficiently sample from prompt candidates. This approach significantly outperforms both human-engineered prompts and several other prompt optimization methods across 11 representative multi-step tasks (an average 10.6\%-29.3\% improvement to current best methods on five LLMs respectively). We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks. Datasets and Codes are available at https://github.com/yongchao98/PROMST. Project Page is available at https://yongchao98.github.io/MIT-REALM-PROMST.

Sunshine Jiang, Xiaolin Fang, Nicholas Roy et al.

Recent advances in diffusion$/$flow-matching policies have enabled imitation learning of complex, multi-modal action trajectories. However, they are computationally expensive because they sample a trajectory of trajectories: a diffusion$/$flow trajectory of action trajectories. They discard intermediate action trajectories, and must wait for the sampling process to complete before any actions can be executed on the robot. We simplify diffusion$/$flow policies by treating action trajectories as flow trajectories. Instead of starting from pure noise, our algorithm samples from a narrow Gaussian around the last action. Then, it incrementally integrates a velocity field learned via flow matching to produce a sequence of actions that constitute a single trajectory. This enables actions to be streamed to the robot on-the-fly during the flow sampling process, and is well-suited for receding horizon policy execution. Despite streaming, our method retains the ability to model multi-modal behavior. We train flows that stabilize around demonstration trajectories to reduce distribution shift and improve imitation learning performance. Streaming flow policy outperforms prior methods while enabling faster policy execution and tighter sensorimotor loops for learning-based robot control. Project website: https://streaming-flow-policy.github.io/

Emily Liu, Michael Noseworthy, Nicholas Roy

In this paper, we investigate a scenario in which a robot learns a low-dimensional representation of a door given a video of the door opening or closing. This representation can be used to infer door-related parameters and predict the outcomes of interacting with the door. Current machine learning based approaches in the doors domain are based primarily on labelled datasets. However, the large quantity of available door data suggests the feasibility of a semisupervised approach based on pretraining. To exploit the hierarchical structure of the dataset where each door has multiple associated images, we pretrain with a structured latent variable model known as a neural statistician. The neural satsitician enforces separation between shared context-level variables (common across all images associated with the same door) and instance-level variables (unique to each individual image). We first demonstrate that the neural statistician is able to learn an embedding that enables reconstruction and sampling of realistic door images. Then, we evaluate the correspondence of the learned embeddings to human-interpretable parameters in a series of supervised inference tasks. It was found that a pretrained neural statistician encoder outperformed analogous context-free baselines when predicting door handedness, size, angle location, and configuration from door images. Finally, in a visual bandit door-opening task with a variety of door configuration, we found that neural statistician embeddings achieve lower regret than context-free baselines.

Vasileios Vasilopoulos, Sebastian Castro, William Vega-Brown et al.

We describe a task and motion planning architecture for highly dynamic systems that combines a domain-independent sampling-based deliberative planning algorithm with a global reactive planner. We leverage the recent development of a reactive, vector field planner that provides guarantees of reachability to large regions of the environment even in the face of unknown or unforeseen obstacles. The reachability guarantees can be formalized using contracts that allow a deliberative planner to reason purely in terms of those contracts and synthesize a plan by choosing a sequence of reactive behaviors and their target configurations, without evaluating specific motion plans between targets. This reduces both the search depth at which plans will be found, and the number of samples required to ensure a plan exists, while crucially preserving correctness guarantees. The result is reduced computational cost of synthesizing plans, and increased robustness of generated plans to actuator noise, model misspecification, or unknown obstacles. Simulation studies show that our hierarchical planning and execution architecture can solve complex navigation and rearrangement tasks, even when faced with narrow passageways or incomplete world information.

Nicholas Roy, Ingmar Posner, Tim Barfoot et al.

Machine learning has long since become a keystone technology, accelerating science and applications in a broad range of domains. Consequently, the notion of applying learning methods to a particular problem set has become an established and valuable modus operandi to advance a particular field. In this article we argue that such an approach does not straightforwardly extended to robotics -- or to embodied intelligence more generally: systems which engage in a purposeful exchange of energy and information with a physical environment. In particular, the purview of embodied intelligent agents extends significantly beyond the typical considerations of main-stream machine learning approaches, which typically (i) do not consider operation under conditions significantly different from those encountered during training; (ii) do not consider the often substantial, long-lasting and potentially safety-critical nature of interactions during learning and deployment; (iii) do not require ready adaptation to novel tasks while at the same time (iv) effectively and efficiently curating and extending their models of the world through targeted and deliberate actions. In reality, therefore, these limitations result in learning-based systems which suffer from many of the same operational shortcomings as more traditional, engineering-based approaches when deployed on a robot outside a well defined, and often narrow operating envelope. Contrary to viewing embodied intelligence as another application domain for machine learning, here we argue that it is in fact a key driver for the advancement of machine learning technology. In this article our goal is to highlight challenges and opportunities that are specific to embodied intelligence and to propose research directions which may significantly advance the state-of-the-art in robot learning.

Matthew Giamou, Filip Marić, David M. Rosen et al.

Inverse kinematics (IK) is the problem of finding robot joint configurations that satisfy constraints on the position or pose of one or more end-effectors. For robots with redundant degrees of freedom, there is often an infinite, nonconvex set of solutions. The IK problem is further complicated when collision avoidance constraints are imposed by obstacles in the workspace. In general, closed-form expressions yielding feasible configurations do not exist, motivating the use of numerical solution methods. However, these approaches rely on local optimization of nonconvex problems, often requiring an accurate initialization or numerous re-initializations to converge to a valid solution. In this work, we first formulate inverse kinematics with complex workspace constraints as a convex feasibility problem whose low-rank feasible points provide exact IK solutions. We then present \texttt{CIDGIK} (Convex Iteration for Distance-Geometric Inverse Kinematics), an algorithm that solves this feasibility problem with a sequence of semidefinite programs whose objectives are designed to encourage low-rank minimizers. Our problem formulation elegantly unifies the configuration space and workspace constraints of a robot: intrinsic robot geometry and obstacle avoidance are both expressed as simple linear matrix equations and inequalities. Our experimental results for a variety of popular manipulator models demonstrate faster and more accurate convergence than a conventional nonlinear optimization-based approach, especially in environments with many obstacles.

Matthew R. Walter, Siddharth Patki, Andrea F. Daniele et al.

Contemporary approaches to perception, planning, estimation, and control have allowed robots to operate robustly as our remote surrogates in uncertain, unstructured environments. This progress now creates an opportunity for robots to operate not only in isolation, but also with and alongside humans in our complex environments. Realizing this opportunity requires an efficient and flexible medium through which humans can communicate with collaborative robots. Natural language provides one such medium, and through significant progress in statistical methods for natural-language understanding, robots are now able to interpret a diverse array of free-form commands. However, most contemporary approaches require a detailed, prior spatial-semantic map of the robot's environment that models the space of possible referents of an utterance. Consequently, these methods fail when robots are deployed in new, previously unknown, or partially-observed environments, particularly when mental models of the environment differ between the human operator and the robot. This paper provides a comprehensive description of a novel learning framework that allows field and service robots to interpret and correctly execute natural-language instructions in a priori unknown, unstructured environments. Integral to our approach is its use of language as a "sensor" -- inferring spatial, topological, and semantic information implicit in the utterance and then exploiting this information to learn a distribution over a latent environment model. We incorporate this distribution in a probabilistic, language grounding model and infer a distribution over a symbolic representation of the robot's action space. We use imitation learning to identify a belief-space policy that reasons over the environment and behavior distributions. We evaluate our framework through a variety navigation and mobile-manipulation experiments.

Christopher Bradley, Adam Pacheck, Gregory J. Stein et al.

We propose a novel planning technique for satisfying tasks specified in temporal logic in partially revealed environments. We define high-level actions derived from the environment and the given task itself, and estimate how each action contributes to progress towards completing the task. As the map is revealed, we estimate the cost and probability of success of each action from images and an encoding of that action using a trained neural network. These estimates guide search for the minimum-expected-cost plan within our model. Our learned model is structured to generalize across environments and task specifications without requiring retraining. We demonstrate an improvement in total cost in both simulated and real-world experiments compared to a heuristic-driven baseline.

Shen Li, Daehyung Park, Yoonchang Sung et al.

We present a task-and-motion planning (TAMP) algorithm robust against a human operator's cooperative or adversarial interventions. Interventions often invalidate the current plan and require replanning on the fly. Replanning can be computationally expensive and often interrupts seamless task execution. We introduce a dynamically reconfigurable planning methodology with behavior tree-based control strategies toward reactive TAMP, which takes the advantage of previous plans and incremental graph search during temporal logic-based reactive synthesis. Our algorithm also shows efficient recovery functionalities that minimize the number of replanning steps. Finally, our algorithm produces a robust, efficient, and complete TAMP solution. Our experimental results show the algorithm results in superior manipulation performance in both simulated and real-world tasks.

Yulun Tian, Katherine Liu, Kyel Ok et al.

We present a multi-robot system for GPS-denied search and rescue under the forest canopy. Forests are particularly challenging environments for collaborative exploration and mapping, in large part due to the existence of severe perceptual aliasing which hinders reliable loop closure detection for mutual localization and map fusion. Our proposed system features unmanned aerial vehicles (UAVs) that perform onboard sensing, estimation, and planning. When communication is available, each UAV transmits compressed tree-based submaps to a central ground station for collaborative simultaneous localization and mapping (CSLAM). To overcome high measurement noise and perceptual aliasing, we use the local configuration of a group of trees as a distinctive feature for robust loop closure detection. Furthermore, we propose a novel procedure based on cycle consistent multiway matching to recover from incorrect pairwise data associations. The returned global data association is guaranteed to be cycle consistent, and is shown to improve both precision and recall compared to the input pairwise associations. The proposed multi-UAV system is validated both in simulation and during real-world collaborative exploration missions at NASA Langley Research Center.

William Vega-Brown, Nicholas Roy

We define an admissibility condition for abstractions expressed using angelic semantics and show that these conditions allow us to accelerate planning while preserving the ability to find the optimal motion plan. We then derive admissible abstractions for two motion planning domains with continuous state. We extract upper and lower bounds on the cost of concrete motion plans using local metric and topological properties of the problem domain. These bounds guide the search for a plan while maintaining performance guarantees. We show that abstraction can dramatically reduce the complexity of search relative to a direct motion planner. Using our abstractions, we find near-optimal motion plans in planning problems involving $10^{13}$ states without using a separate task planner.

Thomas Kollar, Stefanie Tellex, Matthew Walter et al.

Many task domains require robots to interpret and act upon natural language commands which are given by people and which refer to the robot's physical surroundings. Such interpretation is known variously as the symbol grounding problem, grounded semantics and grounded language acquisition. This problem is challenging because people employ diverse vocabulary and grammar, and because robots have substantial uncertainty about the nature and contents of their surroundings, making it difficult to associate the constitutive language elements (principally noun phrases and spatial relations) of the command text to elements of those surroundings. Symbolic models capture linguistic structure but have not scaled successfully to handle the diverse language produced by untrained users. Existing statistical approaches can better handle diversity, but have not to date modeled complex linguistic structure, limiting achievable accuracy. Recent hybrid approaches have addressed limitations in scaling and complexity, but have not effectively associated linguistic and perceptual features. Our framework, called Generalized Grounding Graphs (G^3), addresses these issues by defining a probabilistic graphical model dynamically according to the linguistic parse structure of a natural language command. This approach scales effectively, handles linguistic diversity, and enables the system to associate parts of a command with the specific objects, places, and events in the external world to which they refer. We show that robots can learn word meanings and use those learned meanings to robustly follow natural language commands produced by untrained users. We demonstrate our approach for both mobility commands and mobile manipulation commands involving a variety of semi-autonomous robotic platforms, including a wheelchair, a micro-air vehicle, a forklift, and the Willow Garage PR2.

Genevieve Flaspohler, Nicholas Roy, Yogesh Girdhar

We consider the task of monitoring spatiotemporal phenomena in real-time by deploying limited sampling resources at locations of interest irrevocably and without knowledge of future observations. This task can be modeled as an instance of the classical secretary problem. Although this problem has been studied extensively in theoretical domains, existing algorithms require that data arrive in random order to provide performance guarantees. These algorithms will perform arbitrarily poorly on data streams such as those encountered in robotics and environmental monitoring domains, which tend to have spatiotemporal structure. We focus on the problem of selecting representative samples from phenomena with periodic structure and introduce a novel sample selection algorithm that recovers a near-optimal sample set according to any monotone submodular utility function. We evaluate our algorithm on a seven-year environmental dataset collected at the Martha's Vineyard Coastal Observatory and show that it selects phytoplankton sample locations that are nearly optimal in an information-theoretic sense for predicting phytoplankton concentrations in locations that were not directly sampled. The proposed periodic secretary algorithm can be used with theoretical performance guarantees in many real-time sensing and robotics applications for streaming, irrevocable sample selection from periodic data streams.

Gregory J. Stein, Nicholas Roy

We propose a novel approach for generating high-quality, synthetic data for domain-specific learning tasks, for which training data may not be readily available. We leverage recent progress in image-to-image translation to bridge the gap between simulated and real images, allowing us to generate realistic training data for real-world tasks using only unlabeled real-world images and a simulation. GeneSIS-RT ameliorates the burden of having to collect labeled real-world images and is a promising candidate for generating high-quality, domain-specific, synthetic data. To show the effectiveness of using GeneSIS-RT to create training data, we study two tasks: semantic segmentation and reactive obstacle avoidance. We demonstrate that learning algorithms trained using data generated by GeneSIS-RT make high-accuracy predictions and outperform systems trained on raw simulated data alone, and as well or better than those trained on real data. Finally, we use our data to train a quadcopter to fly 60 meters at speeds up to 3.4 m/s through a cluttered environment, demonstrating that our GeneSIS-RT images can be used to learn to perform mission-critical tasks.

Vasileios Vasilopoulos, William Vega-Brown, Omur Arslan et al.

This paper considers the problem of completing assemblies of passive objects in nonconvex environments, cluttered with convex obstacles of unknown position, shape and size that satisfy a specific separation assumption. A differential drive robot equipped with a gripper and a LIDAR sensor, capable of perceiving its environment only locally, is used to position the passive objects in a desired configuration. The method combines the virtues of a deliberative planner generating high-level, symbolic commands, with the formal guarantees of convergence and obstacle avoidance of a reactive planner that requires little onboard computation and is used online. The validity of the proposed method is verified both with formal proofs and numerical simulations.

Emma Brunskill, Bethany Leffler, Lihong Li et al.

Continuous state spaces and stochastic, switching dynamics characterize a number of rich, realworld domains, such as robot navigation across varying terrain. We describe a reinforcementlearning algorithm for learning in these domains and prove for certain environments the algorithm is probably approximately correct with a sample complexity that scales polynomially with the state-space dimension. Unfortunately, no optimal planning techniques exist in general for such problems; instead we use fitted value iteration to solve the learned MDP, and include the error due to approximate planning in our bounds. Finally, we report an experiment using a robotic car driving over varying terrain to demonstrate that these dynamics representations adequately capture real-world dynamics and that our algorithm can be used to efficiently solve such problems.