Robotic Grasping & Manipulation Planning 2026
Robotic Grasping & Manipulation Planning 2026
The field is transitioning rapidly from classical geometric approaches toward learning-based, data-driven systems. This landscape covers ~60 records spanning patents and literature from 2006 to 2026.
From Grasp Synthesis to Closed-Loop Manipulation
Robotic grasping and manipulation planning integrates computer vision, machine learning, motion planning, contact mechanics, and control theory. The field addresses four interlocking problems: grasp synthesis, motion and trajectory planning, task-oriented manipulation, and closed-loop reactive control — all increasingly handled by learning-based pipelines.
The dataset spans approximately 60 distinct records from 2006 to 2026, covering granted and pending patents across US, WO, and IN jurisdictions alongside peer-reviewed literature. The densest publication cluster falls in 2019–2022, comprising roughly 35 records, reflecting the rapid maturation of deep-learning-based grasp methods and task-aware manipulation.
Commercial IP is concentrated in a small number of large technology companies — Samsung Electronics, Autodesk, and Ocado Innovation — while the majority of technical innovation signals originate from academic institutions globally. This bifurcation between distributed academic research and concentrated commercial IP is a defining structural feature of the landscape.
The 2023–2026 frontier is defined by affordance-based unified pick-and-place networks, shared autonomy frameworks for remote collaboration, and adaptive assembly with dynamic re-grasping. Samsung Electronics holds the most active patent family in the dataset, with four filings across US and WO jurisdictions covering neural-network-integrated task-aware grasp estimation.
Technology Clusters and Publication Trends
The dataset reveals four dominant technical clusters, with learning-based grasp synthesis comprising the largest share (~18–20 records), followed by task-aware grasping, TAMP, and RL-based reactive manipulation. Publication activity peaked in the 2019–2022 window.
Record Count by Technology Cluster — Robotic Grasping Dataset
Learning-based grasp synthesis is the dominant cluster with approximately 18–20 records, reflecting deep-learning method maturation across CNN, point cloud, and generative model approaches.
↗ Click bars to explorePublication Activity by Era — Robotic Grasping Dataset
The 2019–2022 period accounts for approximately 35 of ~60 dataset records, representing the densest publication cluster and peak activity in learning-based and task-aware manipulation research.
↗ Click bars to exploreWhere Grasping and Manipulation Technology Is Being Deployed
The dataset covers six distinct application domains, from industrial assembly and logistics warehousing to space robotics and agricultural harvesting, each with specific technical requirements driving distinct patent and research activity.
Industrial Manufacturing & Assembly
Industrial bin picking and assembly represent the largest application cluster in the dataset. A PLC-integrated deep learning system achieved 95% grasp success at over 350 picks per hour for previously unseen objects. Autodesk’s 2023 patent on adaptive robotic assembly integrates grasp perception, re-grasp determination, and motion planning within a unified robot control application.
Bin Picking & AssemblyLogistics, Warehousing & E-Commerce
Ocado Innovation’s 2021 US patent covers active perception and coordination between robotic vision systems and manipulators, directly targeting warehouse automation. RF-Grasp (2021) enables grasping of fully occluded objects using RFID tags for warehouse environments. REGRAD (2022), a large-scale relational grasp dataset, explicitly targets search-and-grasp in clutter for logistics pipelines.
Warehouse AutomationSpace & Extreme Environment Robotics
On-orbit manipulation is addressed by a 2021 study on robotic grasping of a spent rocket stage, which introduces an Intrinsic Stiffness Matrix-based grasp stability metric for space debris capture. A 2021 survey covers SMS dynamics, contact modeling, and motion planning for on-orbit capture across uncontrolled tumbling objects.
On-Orbit ManipulationAgricultural & Outdoor Robotics
A 2022 study addresses human-inspired grasp planning for fruit and vegetable harvesting for agricultural robots. A 2023 paper describes a dual-arm grape harvesting robot with virtual hand-eye coordination simulation for multi-interaction operation. These works emphasize adapting grasp planners to deformable, irregular natural objects in unstructured outdoor environments.
Agricultural HarvestingCommercial IP Concentration in Robotic Grasping
Among 8 patent records with explicit assignee data, Samsung Electronics leads with 4 filings across US and WO jurisdictions. Commercial IP is concentrated in Samsung, Autodesk, and Ocado Innovation, while academic institutions dominate the broader research literature.
Patent Filings by Named Assignee — Robotic Grasping Dataset
↗ Click bars to exploreSamsung Electronics Co., Ltd.
Samsung Electronics is the single most active patent filer in this dataset, with 4 records spanning US (active), US (pending), and WO jurisdictions filed between 2024 and 2026. All filings relate to “Synergies between pick and place: task-aware grasp estimation,” covering neural network models that jointly process 3D geometry of a target object and placement scene to produce unified grasp and placement affordance information. The multi-jurisdictional strategy across active and pending grants reflects sustained commercial IP positioning for neural-network-integrated robotics.
South Korea / US & WOAutodesk, Inc.
Autodesk holds one pending US patent filed in 2023 titled “Techniques for adaptive robotic assembly,” which integrates grasp perception models, re-grasp determination, and motion planning within a unified robot control application for assembly task execution. The filing targets the industrial assembly domain and specifically addresses dynamic re-grasping within assembly pipelines when initial grasp poses are incompatible with assembly requirements. Patent status is pending as of the dataset coverage date.
United StatesFive Forward-Leaning Trends Shaping 2024–2026
The most recent records in the dataset (2023–2026) identify five emergent directions: affordance-driven unified pick-and-place networks, shared remote collaborative autonomy, relational grasping in clutter, adaptive assembly with re-grasping, and active next-best-view perception.
Affordance-Driven Unified Pick-and-Place Networks
Samsung Electronics’ cluster of US and WO patents (2024–2026) on task-aware grasp estimation represents a move toward single neural network models that jointly determine grasp pose and placement orientation from 3D scene geometry. This eliminates the traditional decoupling of grasping from placement planning, with an active US grant and pending continuations filed through 2026. R&D teams in logistics and service robotics should conduct FTO analysis against this patent family.
Shared and Remote Collaborative Autonomy
Toyota Technological Institute at Chicago’s SHARC framework patent (2024, US, pending) introduces a multi-user remote manipulation architecture with interactive 3D scene understanding for distributed human-robot co-planning. This signals growing interest in cloud-mediated, teleoperation-augmented autonomy for manipulation tasks. The architecture enables multiple remote operators to collaboratively guide robot manipulation in real time.
Learning-Based Grasp Synthesis vs. Task-Aware Grasping
Click any row to explore further.
| Dimension | Learning-Based Grasp Synthesis | Task-Aware Grasping |
|---|---|---|
| Dataset Records | ~18–20 records | ~10–12 records |
| Core Mechanism | CNN / point cloud / generative models predict 6-DoF grasp quality from depth or RGB-D images | Neural networks or embedding models couple grasp selection with downstream task requirements |
| Representative Work | GG-CNN (2018): pixel-wise grasp quality maps from depth at 50Hz | GATER (2022): tool–action–target embeddings achieving 94.6% task inference success |
| Key Metric | 91% physical grasp success on occluded surfaces (2021); real-time 50Hz inference | 94.6% task-specific grasp inference success (GATER, 2022) |
| Training Data | Synthetic and real depth/RGB-D datasets; large-scale simulation (Dex-Net paradigm) | Simulated self-supervision (TOG-Net, 2019); automatically generated task-specific synthetic data (2022) |
| Commercial IP | Concentrated; Samsung Electronics holds active US grants and WO filings (2024–2026) | Also concentrated in Samsung; Autodesk addresses task-aware assembly re-grasping (2023) |
| Primary Application | Bin picking, logistics, e-commerce, cluttered environments | Assembly, tool use, human-robot handover, pick-and-place with placement constraints |
| Maturity Era | 2016–2018 transition; dense activity 2019–2022; still dominant in 2023–2026 | Emerged 2019; peak activity 2022; commercial filings extend to 2026 |
Frequently Asked Questions: Robotic Grasping & Manipulation Patents
Samsung Electronics Co., Ltd. is the single most active patent filer in the dataset, with 4 records covering task-aware grasp estimation across US (active), US (pending), and WO jurisdictions filed between 2024 and 2026. Autodesk, Ocado Innovation, and Toyota Technological Institute at Chicago each hold one patent record.
GG-CNN is a generative convolutional neural network architecture introduced in 2018 that produces pixel-wise grasp quality and orientation maps from depth images at 50Hz, enabling closed-loop grasping at real-time rates. It represents a key milestone in the 2016–2018 transition to deep learning approaches for grasp synthesis.
GATER (2022) models tool–action–target-object relationships in embedding space and achieves 94.6% task-specific grasp inference success. It addresses the coupling between grasp selection and downstream task requirements, going beyond stability-only grasp metrics.
SHARC is a shared autonomy framework for remote collaborative manipulation developed by Toyota Technological Institute at Chicago, filed as a US pending patent in 2024. It introduces a multi-user remote manipulation architecture with interactive 3D scene understanding, enabling distributed human-robot co-planning for manipulation tasks.
US filings dominate, accounting for 5 of 8 patent records with explicit assignee data. WO jurisdiction covers 2 records and IN (India) covers 1 record. The literature corpus is globally distributed with contributions from European, US, and Asian institutions, but commercial IP is concentrated in US and Korean filings.
A 2020 study on meta-reinforcement learning for robotic industrial insertion trains in simulation across a family of insertion tasks, then adapts to new insertion tasks in under 20 real-world trials. This reduces the real-world data requirement to tractable levels for contact-rich industrial insertion applications.
Data and insights on this page are based on a limited patent and literature dataset and are for reference only. Figures may not represent the complete technology landscape.