Approach & Grasp
Locate the randomized needle pose and establish a stable, orientation-aware grasp.
Hierarchical surgical policy learning
Teaching robots the precision of suturing.
A fine-grained surgical robotics intelligence platform that transforms complex suturing into trainable, measurable, reusable skills—combining simulation, expert demonstrations, reinforcement learning and hierarchical control.
PSM–01 ACTIVEDEPTH + RGBPOLICY: INSERT
7-DOFContinuous control
5 SkillsHierarchical sequence
Real timeRobot interaction
The platform
From expert motion to autonomous surgical skill.
The system provides a standardized environment for collecting trajectories, training policies and benchmarking complex robotic suturing. It supports multimodal observations, deformable thread simulation, teleoperation and both reinforcement and imitation learning.
Dual Patient-Side ManipulatorsEndoscopic CameraDeformable Thread PhysicsRGB + Depth + Pose DataGymnasium InterfaceOnline & Offline LearningTeleoperation DataHeadless Sampling
Skill decomposition
One procedure. Five precise policies.
A high-level policy coordinates specialized low-level policies. Each subtask can be trained, evaluated and improved independently, while still operating as part of a complete suturing sequence.
Place
Align the needle tip with the target entry point and prepare the correct approach angle.
Insert
Drive the needle through entry and exit targets while preserving trajectory and orientation.
Handoff
Transfer control of the needle between manipulators with contact-aware precision.
Pullout
Complete extraction and move the needle to a defined end pose for the next stitch.
Live demonstrations
See the policy execute.
These embedded demonstrations show the learned system operating through a complete suturing sequence and the low-level placement skill in the updated simulation environment.
Complete suturing policy
Hierarchical coordination of multiple learned subtasks into one continuous robotic procedure.
Low-level “Place” skill
Needle alignment and positioning demonstrated in the updated surgical simulation workflow.
Control architecture
Intelligence organized like a surgical team.
The high-level controller selects the next task. Specialized low-level policies execute the required motion until their terminal condition is reached, then return control for the next decision.
◎Graspneedle acquisition
⌖Placetarget alignment
↘Insertcontrolled passage
⇄Handoffdual-arm transfer
↗Pulloutfinal extraction
Benchmark intelligence
Performance that can be measured.
Each policy can be evaluated with explicit surgical-task metrics, allowing algorithms to be compared on completion reliability, motion efficiency and time efficiency.
01
Success rate
Proportion of episodes that meet the defined positional, angular and contact criteria.
02
Trajectory length
Total physical travel during a successful episode, exposing unnecessary motion.
03
Time steps
Number of discrete actions required to complete the task successfully.
04
Generalization
Robustness across randomized needle poses, grasp errors and geometric conditions.
Precision meets intelligence.
A presentation-ready vision for safer, repeatable and measurable robotic skill development.
Powered with ❤️ by Solicitous ®
Research and technology presentation. Not intended for direct clinical decision-making or autonomous clinical use without appropriate validation, regulatory review and human oversight.