PoCo: Policy Composition from and for Heterogeneous Robot Learning

Training general robotic policies from heterogeneous data for different tasks is a significant challenge. Existing robotic datasets vary in different modalities such as color, depth, tactile, and proprioceptive information, and collected in different domains such as simulation, real robots, and human videos. Current methods usually collect and pool all data from one domain to train a single policy to handle such heterogeneity in tasks and domains, which is prohibitively expensive and difficult. In this work, we present a flexible approach, dubbed Policy Composition, to combine information across such diverse modalities and domains for learning scene-level and task-level generalized manipulation skills, by composing different data distributions represented with diffusion models. Our method can use task-level composition for multi-task manipulation and be composed with analytic cost functions to adapt policy behaviors at inference time. We train our method on simulation, human, and real robot data and evaluate in tool-use tasks. The composed policy achieves robust and dexterous performance under varying scenes and tasks and outperforms baselines from a single data source in both simulation and real-world experiments. See https://liruiw.github.io/policycomp for more details .

Fleet Policy Learning via Weight Merging and An Application to Robotic Tool-Use

Fleets of robots ingest massive amounts of streaming data generated by interacting with their environments, far more than those that can be stored or transmitted with ease. At the same time, we hope that teams of robots can co-acquire diverse skills through their experiences in varied settings. How can we enable such fleet-level learning without having to transmit or centralize fleet-scale data? In this paper, we investigate distributed learning of policies as a potential solution. To efficiently merge policies in the distributed setting, we propose fleet-merge, an instantiation of distributed learning that accounts for the symmetries that can arise in learning policies that are parameterized by recurrent neural networks. We show that fleet-merge consolidates the behavior of policies trained on 50 tasks in the Meta-World environment, with the merged policy achieving good performance on nearly all training tasks at test time. Moreover, we introduce a novel robotic tool-use benchmark, fleet-tools, for fleet policy learning in compositional and contact-rich robot manipulation tasks, which might be of broader interest, and validate the efficacy of fleet-merge on the benchmark.

GenSim: Generating Robotic Simulation Tasks via Large Language Models

Collecting large amounts of real-world interaction data to train general robotic policies is often prohibitively expensive, thus motivating the use of simulation data. However, existing methods for data generation have generally focused on scene-level diversity (e.g., object instances and poses) rather than task-level diversity, due to the human effort required to come up with and verify novel tasks. This has made it challenging for policies trained on simulation data to demonstrate significant task-level generalization. In this paper, we propose to automatically generate rich simulation environments and expert demonstrations by exploiting a large language models' (LLM) grounding and coding ability. Our approach, dubbed GenSim, has two modes: goal-directed generation, wherein a target task is given to the LLM and the LLM proposes a task curriculum to solve the target task, and exploratory generation, wherein the LLM bootstraps from previous tasks and iteratively proposes novel tasks that would be helpful in solving more complex tasks. We use GPT4 to expand the existing benchmark by ten times to over 100 tasks, on which we conduct supervised finetuning and evaluate several LLMs including finetuned GPTs and Code Llama on code generation for robotic simulation tasks. Furthermore, we observe that LLMs-generated simulation programs can enhance task-level generalization significantly when used for multitask policy training. We further find that with minimal sim-to-real adaptation, the multitask policies pretrained on GPT4-generated simulation tasks exhibit stronger transfer to unseen long-horizon tasks in the real world and outperform baselines by 25%. See the project website (https://liruiw.github.io/gensim) for code, demos, and videos.

3D Neural Embedding Likelihood for Robust Sim-to-Real Transfer in Inverse Graphics

A central challenge in 3D scene perception via inverse graphics is robustly modeling the gap between 3D graphics and real-world data. We propose a novel 3D Neural Embedding Likelihood (3DNEL) over RGB-D images to address this gap. 3DNEL uses neural embeddings to predict 2D-3D correspondences from RGB and combines this with depth in a principled manner. 3DNEL is trained entirely from synthetic images and generalizes to real-world data. To showcase this capability, we develop a multi-stage inverse graphics pipeline that uses 3DNEL for 6D object pose estimation from real RGB-D images. Our method outperforms the previous state-of-the-art in sim-to-real pose estimation on the YCB-Video dataset, and improves robustness, with significantly fewer large-error predictions. Unlike existing bottom-up, discriminative approaches that are specialized for pose estimation, 3DNEL adopts a probabilistic generative formulation that jointly models multi-object scenes. This generative formulation enables easy extension of 3DNEL to additional tasks like object and camera tracking from video, using principled inference in the same probabilistic model without task specific retraining.

NeRF in the Palm of Your Hand: Corrective Augmentation for Robotics via Novel-View Synthesis

Expert demonstrations are a rich source of supervision for training visual robotic manipulation policies, but imitation learning methods often require either a large number of demonstrations or expensive online expert supervision to learn reactive closed-loop behaviors. In this work, we introduce SPARTN (Synthetic Perturbations for Augmenting Robot Trajectories via NeRF): a fully-offline data augmentation scheme for improving robot policies that use eye-in-hand cameras. Our approach leverages neural radiance fields (NeRFs) to synthetically inject corrective noise into visual demonstrations, using NeRFs to generate perturbed viewpoints while simultaneously calculating the corrective actions. This requires no additional expert supervision or environment interaction, and distills the geometric information in NeRFs into a real-time reactive RGB-only policy. In a simulated 6-DoF visual grasping benchmark, SPARTN improves success rates by 2.8$\times$ over imitation learning without the corrective augmentations and even outperforms some methods that use online supervision. It additionally closes the gap between RGB-only and RGB-D success rates, eliminating the previous need for depth sensors. In real-world 6-DoF robotic grasping experiments from limited human demonstrations, our method improves absolute success rates by $22.5\%$ on average, including objects that are traditionally challenging for depth-based methods. See video results at \url{https://bland.website/spartn}.

PredictionNet: Real-Time Joint Probabilistic Traffic Prediction for Planning, Control, and Simulation

Predicting the future motion of traffic agents is crucial for safe and efficient autonomous driving. To this end, we present PredictionNet, a deep neural network (DNN) that predicts the motion of all surrounding traffic agents together with the ego-vehicle's motion. All predictions are probabilistic and are represented in a simple top-down rasterization that allows an arbitrary number of agents. Conditioned on a multilayer map with lane information, the network outputs future positions, velocities, and backtrace vectors jointly for all agents including the ego-vehicle in a single pass. Trajectories are then extracted from the output. The network can be used to simulate realistic traffic, and it produces competitive results on popular benchmarks. More importantly, it has been used to successfully control a real-world vehicle for hundreds of kilometers, by combining it with a motion planning/control subsystem. The network runs faster than real-time on an embedded GPU, and the system shows good generalization (across sensory modalities and locations) due to the choice of input representation. Furthermore, we demonstrate that by extending the DNN with reinforcement learning (RL), it can better handle rare or unsafe events like aggressive maneuvers and crashes.

Hierarchical Policies for Cluttered-Scene Grasping with Latent Plans

6D grasping in cluttered scenes is a longstanding robotic manipulation problem. Open-loop manipulation pipelines can fail due to modularity and error sensitivity while most end-to-end grasping policies with raw perception inputs have not yet scaled to complex scenes with obstacles. In this work, we propose a new method to close the gap through sampling and selecting plans in the latent space. Our hierarchical framework learns collision-free target-driven grasping based on partial point cloud observations. Our method learns an embedding space to represent expert grasping plans and a variational autoencoder to sample diverse latent plans at inference time. Furthermore, we train a latent plan critic for plan selection and an option classifier for switching to an instance grasping policy through hierarchical reinforcement learning. We evaluate and analyze our method and compare against several baselines in simulation, and demonstrate that the latent planning can generalize to the real-world cluttered-scene grasping task. Our videos and code can be found at https://sites.google.com/view/latent-grasping .

Goal-Auxiliary Actor-Critic for 6D Robotic Grasping with Point Clouds

6D robotic grasping beyond top-down bin-picking scenarios is a challenging task. Previous solutions based on 6D grasp synthesis with robot motion planning usually operate in an open-loop setting without considering the dynamics and contacts of objects, which makes them sensitive to grasp synthesis errors. In this work, we propose a novel method for learning closed-loop control policies for 6D robotic grasping using point clouds from an egocentric camera. We combine imitation learning and reinforcement learning in order to grasp unseen objects and handle the continuous 6D action space, where expert demonstrations are obtained from a joint motion and grasp planner. We introduce a goal-auxiliary actor-critic algorithm, which uses grasping goal prediction as an auxiliary task to facilitate policy learning. The supervision on grasping goals can be obtained from the expert planner for known objects or from hindsight goals for unknown objects. Overall, our learned closed-loop policy achieves over 90% success rates on grasping various ShapeNet objects and YCB objects in the simulation. Our video can be found at https://www.youtube.com/watch?v=rKsCRXLykiY&t=1s .

Manipulation Trajectory Optimization with Online Grasp Synthesis and Selection

In robot manipulation, planning the motion of a robot manipulator to grasp an object is a fundamental problem. A manipulation planner needs to generate a trajectory of the manipulator arm to avoid obstacles in the environment and plan an end-effector pose for grasping. While trajectory planning and grasp planning are often tackled separately, how to efficiently integrate the two planning problems remains a challenge. In this work, we present a novel method for joint motion and grasp planning. Our method integrates manipulation trajectory optimization with online grasp synthesis and selection, where we apply online learning techniques to select goal configurations for grasping, and introduce a new grasp synthesis algorithm to generate grasps online. We evaluate our planning approach and demonstrate that our method generates robust and efficient motion plans for grasping in cluttered scenes.