Despite great strides in language-guided manipulation, existing work has been constrained to table-top settings. Table-tops allow for perfect and consistent camera angles, properties are that do not hold in mobile manipulation. Task plans that involve moving around the environment must be robust to egocentric views and changes in the plane and angle of grasp. A further challenge is ensuring this is all true while still being able to learn skills efficiently from limited data. We propose Spatial-Language Attention Policies (SLAP) as a solution. SLAP uses three-dimensional tokens as the input representation to train a single multi-task, language-conditioned action prediction policy. Our method shows an 80% success rate in the real world across eight tasks with a single model, and a 47.5% success rate when unseen clutter and unseen object configurations are introduced, even with only a handful of examples per task. This represents an improvement of 30% over prior work (20% given unseen distractors and configurations). We see a 4x improvement over baseline in mobile manipulation setting. In addition, we show how SLAPs robustness allows us to execute Task Plans from open-vocabulary instructions using a large language model for multi-step mobile manipulation. For videos, see the website: https://robotslap.github.io

Manipulating objects without grasping them is an essential component of human dexterity, referred to as non-prehensile manipulation. Non-prehensile manipulation may enable more complex interactions with the objects, but also presents challenges in reasoning about gripper-object interactions. In this work, we introduce Hybrid Actor-Critic Maps for Manipulation (HACMan), a reinforcement learning approach for 6D non-prehensile manipulation of objects using point cloud observations. HACMan proposes a temporally-abstracted and spatially-grounded object-centric action representation that consists of selecting a contact location from the object point cloud and a set of motion parameters describing how the robot will move after making contact. We modify an existing off-policy RL algorithm to learn in this hybrid discrete-continuous action representation. We evaluate HACMan on a 6D object pose alignment task in both simulation and in the real world. On the hardest version of our task, with randomized initial poses, randomized 6D goals, and diverse object categories, our policy demonstrates strong generalization to unseen object categories without a performance drop, achieving an 89% success rate on unseen objects in simulation and 50% success rate with zero-shot transfer in the real world. Compared to alternative action representations, HACMan achieves a success rate more than three times higher than the best baseline. With zero-shot sim2real transfer, our policy can successfully manipulate unseen objects in the real world for challenging non-planar goals, using dynamic and contact-rich non-prehensile skills. Videos can be found on the project website: https://hacman-2023.github.io.

Multi-object rearrangement is a crucial skill for service robots, and commonsense reasoning is frequently needed in this process. However, achieving commonsense arrangements requires knowledge about objects, which is hard to transfer to robots. Large language models (LLMs) are one potential source of this knowledge, but they do not naively capture information about plausible physical arrangements of the world. We propose LLM-GROP, which uses prompting to extract commonsense knowledge about semantically valid object configurations from an LLM and instantiates them with a task and motion planner in order to generalize to varying scene geometry. LLM-GROP allows us to go from natural-language commands to human-aligned object rearrangement in varied environments. Based on human evaluations, our approach achieves the highest rating while outperforming competitive baselines in terms of success rate while maintaining comparable cumulative action costs. Finally, we demonstrate a practical implementation of LLM-GROP on a mobile manipulator in real-world scenarios. Supplementary materials are available at: https://sites.google.com/view/llm-grop

Therefore, we split the action prediction problem into two steps: (1) we predict a Most Relevant Point, or MRP, which tells us which region of the world the policy must attend to; and (2) we reactively predict actions which determine where the robot should move in relation to that MRP: for example, how to approach the handle of an oven and when to close the gripper to grasp it. These two operations are performed sequentially using a modified PointNet++ (Qi et al., 2017) model that we refer to as Attention-based PointNet (A-PointNet), shown in Figure 2. The MRP Predictor can then be agnostic to the position of the robot, instead focusing on the features of the object relevant to the overall task, while the Action Predictor can learn to focus on features relevant just to what the next action should be. For example, in Figure 7, the MRP Predictor learns to focus on the handle; the Action Predictor focuses on the angle of the oven door. Image Pre-Processing First we convert the RGB and depth images into a point cloud. We augment the point cloud of the current timestep with our context c, the point cloud from the beginning of the episode. This aids both in combating occlusion, as well as in disambiguating between similar observations that occur during different trajectories. To reduce compute, we crop the working area to 1m, and down-sample using grid pooling, with a resolution of 1cm for the current timestep and 2.5cm for the context. Specifically, we select a random point in each voxel, to reduce overfitting.

We propose CLIP-Fields, an implicit scene model that can be used for a variety of tasks, such as segmentation, instance identification, semantic search over space, and view localization. CLIP-Fields learns a mapping from spatial locations to semantic embedding vectors. Importantly, we show that this mapping can be trained with supervision coming only from web-image and web-text trained models such as CLIP, Detic, and Sentence-BERT; and thus uses no direct human supervision. When compared to baselines like Mask-RCNN, our method outperforms on few-shot instance identification or semantic segmentation on the HM3D dataset with only a fraction of the examples. Finally, we show that using CLIP-Fields as a scene memory, robots can perform semantic navigation in real-world environments. Our code and demonstration videos are available here: https://mahis.life/clip-fields

Robots operating in human environments must be able to rearrange objects into semantically-meaningful configurations, even if these objects are previously unseen. In this work, we focus on the problem of building physically-valid structures without step-by-step instructions. We propose StructDiffusion, which combines a diffusion model and an object-centric transformer to construct structures given partial-view point clouds and high-level language goals, such as "set the table". Our method can perform multiple challenging language-conditioned multi-step 3D planning tasks using one model. StructDiffusion even improves the success rate of assembling physically-valid structures out of unseen objects by on average 16% over an existing multi-modal transformer model trained on specific structures. We show experiments on held-out objects in both simulation and on real-world rearrangement tasks. Importantly, we show how integrating both a diffusion model and a collision-discriminator model allows for improved generalization over other methods when rearranging previously-unseen objects. For videos and additional results, see our website: https://structdiffusion.github.io/.

In order for robots to follow open-ended instructions like "go open the brown cabinet over the sink", they require an understanding of both the scene geometry and the semantics of their environment. Robotic systems often handle these through separate pipelines, sometimes using very different representation spaces, which can be suboptimal when the two objectives conflict. In this work, we present USA-Net, a simple method for constructing a world representation that encodes both the semantics and spatial affordances of a scene in a differentiable map. This allows us to build a gradient-based planner which can navigate to locations in the scene specified using open-ended vocabulary. We use this planner to consistently generate trajectories which are both shorter 5-10% shorter and 10-30% closer to our goal query in CLIP embedding space than paths from comparable grid-based planners which don't leverage gradient information. To our knowledge, this is the first end-to-end differentiable planner optimizes for both semantics and affordance in a single implicit map. Code and visuals are available at our website: https://usa.bolte.cc/

Images are a convenient way to specify which particular object instance an embodied agent should navigate to. Solving this task requires semantic visual reasoning and exploration of unknown environments. We present a system that can perform this task in both simulation and the real world. Our modular method solves sub-tasks of exploration, goal instance re-identification, goal localization, and local navigation. We re-identify the goal instance in egocentric vision using feature-matching and localize the goal instance by projecting matched features to a map. Each sub-task is solved using off-the-shelf components requiring zero fine-tuning. On the HM3D InstanceImageNav benchmark, this system outperforms a baseline end-to-end RL policy 7x and a state-of-the-art ImageNav model 2.3x (56% vs 25% success). We deploy this system to a mobile robot platform and demonstrate effective real-world performance, achieving an 88% success rate across a home and an office environment.

Rearrangement tasks have been identified as a crucial challenge for intelligent robotic manipulation, but few methods allow for precise construction of unseen structures. We propose a visual foresight model for pick-and-place rearrangement manipulation which is able to learn efficiently. In addition, we develop a multi-modal action proposal module which builds on the Goal-Conditioned Transporter Network, a state-of-the-art imitation learning method. Our image-based task planning method, Transporters with Visual Foresight, is able to learn from only a handful of data and generalize to multiple unseen tasks in a zero-shot manner. TVF is able to improve the performance of a state-of-the-art imitation learning method on unseen tasks in simulation and real robot experiments. In particular, the average success rate on unseen tasks improves from 55.4% to 78.5% in simulation experiments and from 30% to 63.3% in real robot experiments when given only tens of expert demonstrations. Video and code are available on our project website: https://chirikjianlab.github.io/tvf/

Robots need to be able to learn concepts from their users in order to adapt their capabilities to each user's unique task. But when the robot operates on high-dimensional inputs, like images or point clouds, this is impractical: the robot needs an unrealistic amount of human effort to learn the new concept. To address this challenge, we propose a new approach whereby the robot learns a low-dimensional variant of the concept and uses it to generate a larger data set for learning the concept in the high-dimensional space. This lets it take advantage of semantically meaningful privileged information only accessible at training time, like object poses and bounding boxes, that allows for richer human interaction to speed up learning. We evaluate our approach by learning prepositional concepts that describe object state or multi-object relationships, like above, near, or aligned, which are key to user specification of task goals and execution constraints for robots. Using a simulated human, we show that our approach improves sample complexity when compared to learning concepts directly in the high-dimensional space. We also demonstrate the utility of the learned concepts in motion planning tasks on a 7-DoF Franka Panda robot.