In this episode of Here’s an Idea, we speak to three researchers who are finding ways to automate surgical tasks, from suturing, to spotting tumors, to operating one of the biggest machines in surgery today: The Da Vinci.
Episode Highlights:
- (1:06) What is the Da Vinci?
- (4:25) A 'Superhuman' Robotic Test: The Peg Transfer
- (12:04) Meet 'STAR,' A Suturing Robot
- (16:28) Robots, Guided by MRI
Subscribe or listen via your preferred podcast provider here.
Testing the Da Vinci, with 'Superhuman' Speed: The Da Vinci system, fixed with instruments and a stereo camera. allows a trained surgeon to sit behind a console and remotely control the tools with a joystick and foot pedals. Ken Goldberg and his team at UC Berkeley want to automate some of the Da Vinci's functions.
In part 1 of our episode, Goldberg explains how he and the students got the Da Vinci to autonomously perform a training exercise — the peg transfer test — with "superhuman" speed and accuracy.
Learn more about Ken Goldberg and his lab and research at UC Berkeley . (Read 50 papers on surgical robotics.)
A 'STAR' Suturer: While robots won't be performing full surgeries anytime soon, an operation's tedious subtasks could use some automation. Take, suturing, for example. Axel Krieger from Johns Hopkins created a technology called the Smart Tissue Autonomous Robot . STAR automates the process of suturing together two segments of an intestine after a portion of the organ is removed.
What makes the STAR so "smart" exactly? Part 2 of our episode reviews how the robot follows carefully placed infrared biomarkers.
MRI-Guided Robots: In 2015, Greg Fischer, along with fellow researchers at Worcester Polytechnic Institute, built a robot that finds its way through a patient to potentially dangerous tissue, using real-time images from an MRI as a navigational guide . Looking at real-time MRI images, the doctor can identify parts of the prostate, for example, that appear suspicious, and direct the robotic tool – specifically the needle of the robotic tool — to those spots for imaging or biopsies.
What do you think of the episode?
Subscribe to the Here's an Idea newsletter.
Transcript
00:00:07 the covid-19 pandemic has caused use of e-commerce to Skyrocket while making it unsafe for warehouse workers to fulfill orders in close proximity these two effects have dramatically increased demand for robots in warehouses and distribution centers robots are currently much slower than human workers matching human efficiency with a robot requires Advan
00:00:30 es in sensing motion planning and robot Hardware recent advances in sensing and robot Hardware are closing the Gap and motion planning remains a bottleneck our grasp optimized motion planner Gump optimizes our motions by solving a sequence of quadratic problems Gump simultaneously optimizes the grasp angle to discover even faster trajectories
00:01:01 however the resulting high-s speed motions can be difficult for the robot to execute resulting in overshoot and wear down robot joints to address this we were able to compute minimum jerk motions but we found this required so much additional computation that it was not practical in response we explored recent advances in deep learning
00:01:31 we computed thousands of smooth motions and used them as examples to train a two-stage deep neural network where the first stage proposes trajectories of varying length and the second stage selects the optimal one we then use this to warm start the optimal planner resulting in a smooth final
00:01:54 trajectory to evaluate the results We compare the original motion planner with a deep learning Planner on on a thousand random trajectories the original planner requires 29 seconds per trajectory while the Deep learning planner only requires 80 milliseconds a 300 times speed up this unique combination of fast and smooth robot motion planning fills an important Gap to enable robots to be
00:02:21 practical for industrial packing and package handling
