Consider the scenario where a human cleans a table and a robot observing the scene is instructed with the task "Remove the cloth using which I wiped the table". Instruction following with temporal reasoning requires the robot to identify the relevant past object interaction, ground the object of interest in the present scene, and execute the task according to the human's instruction. Directly grounding utterances referencing past interactions to grounded objects is challenging due to the multi-hop nature of references to past interactions and large space of object groundings in a video stream observing the robot's workspace. Our key insight is to factor the temporal reasoning task as (i) estimating the video interval associated with event reference, (ii) performing spatial reasoning over the interaction frames to infer the intended object (iii) semantically track the object's location till the current scene to enable future robot interactions. Our approach leverages existing large pre-trained models (which possess inherent generalization capabilities) and combines them appropriately for temporal grounding tasks. Evaluation on a video-language corpus acquired with a robot manipulator displaying rich temporal interactions in spatially-complex scenes displays an average accuracy of 70.10%. The dataset, code, and videos are available at https://reail-iitdelhi.github.io/temporalreasoning.github.io/ .

Automatically detecting and recovering from failures is an important but challenging problem for autonomous robots. Most of the recent work on learning to plan from demonstrations lacks the ability to detect and recover from errors in the absence of an explicit state representation and/or a (sub-) goal check function. We propose an approach (blending learning with symbolic search) for automated error discovery and recovery, without needing annotated data of failures. Central to our approach is a neuro-symbolic state representation, in the form of dense scene graph, structured based on the objects present within the environment. This enables efficient learning of the transition function and a discriminator that not only identifies failures but also localizes them facilitating fast re-planning via computation of heuristic distance function. We also present an anytime version of our algorithm, where instead of recovering to the last correct state, we search for a sub-goal in the original plan minimizing the total distance to the goal given a re-planning budget. Experiments on a physics simulator with a variety of simulated failures show the effectiveness of our approach compared to existing baselines, both in terms of efficiency as well as accuracy of our recovery mechanism.

Our goal is to enable embodied agents to learn inductively generalizable spatial concepts, e.g., learning staircase as an inductive composition of towers of increasing height. Given a human demonstration, we seek a learning architecture that infers a succinct ${program}$ representation that explains the observed instance. Additionally, the approach should generalize inductively to novel structures of different sizes or complex structures expressed as a hierarchical composition of previously learned concepts. Existing approaches that use code generation capabilities of pre-trained large (visual) language models, as well as purely neural models, show poor generalization to a-priori unseen complex concepts. Our key insight is to factor inductive concept learning as (i) ${\it Sketch:}$ detecting and inferring a coarse signature of a new concept (ii) ${\it Plan:}$ performing MCTS search over grounded action sequences (iii) ${\it Generalize:}$ abstracting out grounded plans as inductive programs. Our pipeline facilitates generalization and modular reuse, enabling continual concept learning. Our approach combines the benefits of the code generation ability of large language models (LLM) along with grounded neural representations, resulting in neuro-symbolic programs that show stronger inductive generalization on the task of constructing complex structures in relation to LLM-only and neural-only approaches. Furthermore, we demonstrate reasoning and planning capabilities with learned concepts for embodied instruction following.

Given the task of positioning a ball-like object to a goal region beyond direct reach, humans can often throw, slide, or rebound objects against the wall to attain the goal. However, enabling robots to reason similarly is non-trivial. Existing methods for physical reasoning are data-hungry and struggle with complexity and uncertainty inherent in the real world. This paper presents PhyPlan, a novel physics-informed planning framework that combines physics-informed neural networks (PINNs) with modified Monte Carlo Tree Search (MCTS) to enable embodied agents to perform dynamic physical tasks. PhyPlan leverages PINNs to simulate and predict outcomes of actions in a fast and accurate manner and uses MCTS for planning. It dynamically determines whether to consult a PINN-based simulator (coarse but fast) or engage directly with the actual environment (fine but slow) to determine optimal policy. Given an unseen task, PhyPlan can infer the sequence of actions and learn the latent parameters, resulting in a generalizable approach that can rapidly learn to perform novel physical tasks. Evaluation with robots in simulated 3D environments demonstrates the ability of our approach to solve 3D-physical reasoning tasks involving the composition of dynamic skills. Quantitatively, PhyPlan excels in several aspects: (i) it achieves lower regret when learning novel tasks compared to the state-of-the-art, (ii) it expedites skill learning and enhances the speed of physical reasoning, (iii) it demonstrates higher data efficiency compared to a physics un-informed approach.

Manipulating unseen objects is challenging without a 3D representation, as objects generally have occluded surfaces. This requires physical interaction with objects to build their internal representations. This paper presents an approach that enables a robot to rapidly learn the complete 3D model of a given object for manipulation in unfamiliar orientations. We use an ensemble of partially constructed NeRF models to quantify model uncertainty to determine the next action (a visual or re-orientation action) by optimizing informativeness and feasibility. Further, our approach determines when and how to grasp and re-orient an object given its partial NeRF model and re-estimates the object pose to rectify misalignments introduced during the interaction. Experiments with a simulated Franka Emika Robot Manipulator operating in a tabletop environment with benchmark objects demonstrate an improvement of (i) 14% in visual reconstruction quality (PSNR), (ii) 20% in the geometric/depth reconstruction of the object surface (F-score) and (iii) 71% in the task success rate of manipulating objects a-priori unseen orientations/stable configurations in the scene; over current methods. The project page can be found here: https://actnerf.github.io.

Given the task of positioning a ball-like object to a goal region beyond direct reach, humans can often throw, slide, or rebound objects against the wall to attain the goal. However, enabling robots to reason similarly is non-trivial. Existing methods for physical reasoning are data-hungry and struggle with complexity and uncertainty inherent in the real world. This paper presents PhyPlan, a novel physics-informed planning framework that combines physics-informed neural networks (PINNs) with modified Monte Carlo Tree Search (MCTS) to enable embodied agents to perform dynamic physical tasks. PhyPlan leverages PINNs to simulate and predict outcomes of actions in a fast and accurate manner and uses MCTS for planning. It dynamically determines whether to consult a PINN-based simulator (coarse but fast) or engage directly with the actual environment (fine but slow) to determine optimal policy. Evaluation with robots in simulated 3D environments demonstrates the ability of our approach to solve 3D-physical reasoning tasks involving the composition of dynamic skills. Quantitatively, PhyPlan excels in several aspects: (i) it achieves lower regret when learning novel tasks compared to state-of-the-art, (ii) it expedites skill learning and enhances the speed of physical reasoning, (iii) it demonstrates higher data efficiency compared to a physics un-informed approach.

Given a natural language instruction and an input scene, our goal is to train a model to output a manipulation program that can be executed by the robot. Prior approaches for this task possess one of the following limitations: (i) rely on hand-coded symbols for concepts limiting generalization beyond those seen during training [1] (ii) infer action sequences from instructions but require dense sub-goal supervision [2] or (iii) lack semantics required for deeper object-centric reasoning inherent in interpreting complex instructions [3]. In contrast, our approach can handle linguistic as well as perceptual variations, end-to-end trainable and requires no intermediate supervision. The proposed model uses symbolic reasoning constructs that operate on a latent neural object-centric representation, allowing for deeper reasoning over the input scene. Central to our approach is a modular structure consisting of a hierarchical instruction parser and an action simulator to learn disentangled action representations. Our experiments on a simulated environment with a 7-DOF manipulator, consisting of instructions with varying number of steps and scenes with different number of objects, demonstrate that our model is robust to such variations and significantly outperforms baselines, particularly in the generalization settings. The code, dataset and experiment videos are available at https://nsrmp.github.io

Robots assisting us in environments such as factories or homes must learn to make use of objects as tools to perform tasks, for instance, using a tray to carry objects. We consider the problem of learning common sense knowledge of when a tool may be useful and how its use may be composed with other tools to accomplish a high-level task instructed by a human. Specifically, we introduce a novel neural model, termed TOOLTANGO, that first predicts the next tool to be used, and then uses this information to predict the next action. We show that this joint model can inform learning of a fine-grained policy enabling the robot to use a particular tool in sequence and adds a significant value in making the model more accurate. TOOLTANGO encodes the world state, comprising objects and symbolic relationships between them, using a graph neural network and is trained using demonstrations from human teachers instructing a virtual robot in a physics simulator. The model learns to attend over the scene using knowledge of the goal and the action history, finally decoding the symbolic action to execute. Crucially, we address generalization to unseen environments where some known tools are missing, but unseen alternative tools are present. We show that by augmenting the representation of the environment with pre-trained embeddings derived from a knowledge-base, the model can generalize effectively to novel environments. Experimental results show at least 48.8-58.1% absolute improvement over the baselines in predicting successful symbolic plans for a simulated mobile manipulator in novel environments with unseen objects. This work takes a step in the direction of enabling robots to rapidly synthesize robust plans for complex tasks, particularly in novel settings.

Robots assisting us in factories or homes must learn to make use of objects as tools to perform tasks, e.g., a tray for carrying objects. We consider the problem of learning commonsense knowledge of when a tool may be useful and how its use may be composed with other tools to accomplish a high-level task instructed by a human. We introduce TANGO, a novel neural model for predicting task-specific tool interactions. TANGO is trained using demonstrations obtained from human teachers instructing a virtual robot in a physics simulator. TANGO encodes the world state comprising of objects and symbolic relationships between them using a graph neural network. The model learns to attend over the scene using knowledge of the goal and the action history, finally decoding the symbolic action to execute. Crucially, we address generalization to unseen environments where some known tools are missing, but alternative unseen tools are present. We show that by augmenting the representation of the environment with pre-trained embeddings derived from a knowledge-base, the model can generalize effectively to novel environments. Experimental results show a 60.5-78.9% improvement over the baseline in predicting successful symbolic plans in unseen settings for a simulated mobile manipulator.

In this paper, we address the problem of continually parsing a stream of 3D point cloud data acquired from a laser sensor mounted on a road vehicle. We leverage an online star clustering algorithm coupled with an incremental belief update in an evolving undirected graphical model. The fusion of these techniques allows the robot to parse streamed data and to continually improve its understanding of the world. The core competency produced is an ability to infer object classes from similarities based on appearance and shape features, and to concurrently combine that with a spatial smoothing algorithm incorporating geometric consistency. This formulation of feature-space star clustering modulating the potentials of a spatial graphical model is entirely novel. In our method, the two sources of information: feature similarity and geometrical consistency are fed continu- ally into the system, improving the belief over the class distributions as new data arrives. The algorithm obviates the need for hand-labeled training data and makes no apriori assumptions on the number or characteristics of object categories. Rather, they are learnt incrementally over time from streamed input data. In experiments per- formed on real 3D laser data from an outdoor scene, we show that our approach is capable of obtaining an ever- improving unsupervised scene categorization.