Could Human-aware Algorithms Be the Key to Motion Planning in Robotics?
Making robots more human-like could be a necessary hurdle to cross, especially concerning motion control and planning. What are places like MIT and Mitsubishi trying to accomplish with the right algorithms?
There has always been a fine, gray line for making robots too lifelike. However, if done correctly and with the right aspects of humanity (dexterity, motion, sensing, etc.), robotics could overcome some significant hurdles.
Recently, lead researchers at MIT’s Computer Science & Artificial Intelligence Laboratory (CSAIL) have developed a system that can manipulate the design, simulate the robot doing a task, and provide an optimization score to assess control.
Rigid robots can be re-optimized into free moving tools by reapplying CSAIL’s latest algorithm that increases flexibility. Image used courtesy of MIT’s CSAIL
Contact aware robotic designs aim to make robots more human-like. Though advanced dexterity in humans is not an easy trait to replicate, MIT researchers, as well as other companies and teams, have continued to improve movements in robots.
Today, this article will look at CSAIL's research and what they are doing to further the field of robotics, as well as take a look at some other similar advancements that are taking place.
The Impact of CSAIL’s Research
Getting started, let's establish an overarching conversation about CSAIL's research and its impact.
Typically, dexterity for robotics often comes in the form of grippers. Observing standard robot grippers, they are frequently optimized for control with an existing, fixed design and are not equipped to handle non-defined obstacles or tasks.
For example, a robotic arm extends outward, grabs an object, and retracts it back. All actions would be predefined, and the designer would create a fixed algorithm to accomplish these tasks in a repeatable fashion. However, this design would not handle turning over a box, similar to what our arms and hands can do if we chose to. This 'new' movement would require an entirely different design; a new algorithm would be required, thus making robotic devices challenging to control and establish in unstructured environments.
Let's quickly compare human characteristics such as perception, decision-making, and actions/movements to robots. These characteristics or wants are easy to discuss but challenging to design.
A primary way for humans to perceive objects is through vision. On the other hand, robots require cameras and LiDAR (or any other optical sensors). Another added attribute to humans is how they can also use sounds to perceive their surroundings; however, robots require various microphones, sonar, and sensors to replicate these human abilities.
An example of some sensors being used in a robotic arm. Image used courtesy of Kerzel et al
In regards to decision-making, complex software is needed to allow robots to analyze data in real-time. This software can lead to decisive movements and actions to be completed. Unlike humans, robots can be programmed to repeat a movement consecutively but struggle to change between several different tasks.
When it comes to CSAIL, its design is innovative in the sense that it can be perceived closer to that of a human and in real-time.
The researchers developed cage-based deformation, a classic geometry processing technique used in computer graphics to deform a high-resolution mech in real-time.
CSAIL's cage deformation claims to allow for a more complex and natural design. Screenshot used courtesy of Zu et al
The cage-based deformation method is applied to the shape of each robot component by positions of cage handles, which are defined by the users based on commands.
Adding cage handles increases the degrees of freedom for the robot. The designer could then program this gradient-like optimization into any system to increase the flexibility of robotic components in any workspace.
Adding to the idea of optimized programming, let's take a look at algorithms for path planning.
Breaking-down 3D Path Planning
Three-dimensional (3D) path planning for incoming targets was surveyed in 2016 by a lead engineer, Liang Yang, at Shenyang Institute of Automation in China. He found optimal-collision-free paths in a 3D workspace.
Path planning aims to advance robotics through algorithms that can become a standard for aerial, ground, and underwater robots. In this academic paper, a two-step procedure of 3D path planning was composed to generate a smooth path of action. The first step allows the robot to perceive its environment, modeling a grid map of current surroundings. The second step deployed the path planning algorithm to determine the best route in a time-efficient manner.
Flowchart of 3D path planning, each algorithm will complete simultaneously and are not dependent on each other. The overall software will collect data from each section and decide which path to follow. Image used courtesy of Yang et al
A two-step procedure might sound simple enough; however, these two steps can be broken up into more than four, depending on the given tasks. These steps can be split into sampling, optimal, mathematical, bioinspired, and multi-fusion algorithms to handle various environments.
By gathering large amounts of data in real-time, sampling is at the start of the path planning process, leading to nodal/optimal elements. These nodal/optimal elements allow robots to analyze the data from point to point. The robot will model the environment completely from that node configuration to choose the most appropriate route to follow, free of hurdles or potential collisions.
What makes the last algorithm crucial is when the task to move forward and grab an object comes into play, and there is no single approach for the robot to choose from. This need is where multi-fusion-based algorithms thrive; facing unknown environments, the robot can achieve the necessary task or movement by utilizing sensors and the data previously collected.
Allowing robots to achieve the ability to have multiple functions and to decide which movements are necessary, is an enormous hurdle. Though this algorithm sounds like it is a step in the right direction to help overcome these limitations, Mitsubishi is also developing algorithms to improve robotics.
Collision Avoidance for Cobots
Collaborative robots (cobots) are repeatable, reliable, efficient, and cost-effective technology in manufacturing plants for industrial and automotive applications. These cobots provide a boost in production lines and can be added alongside working human counterparts. Some rising cases limit cobots, namely flexibility and compactness.
Mitsubishi Electric, a Japanese multinational electronics and electrical equipment manufacturer, have developed ways for cobots to control speed and flexibility in various workspaces while maintaining design sophistication.
Mitsubishi's real-time cobot. Image used courtesy of Mitsubishi Electric
Programmed cobots can follow predefined motions and actions without flexibility due to having no predefined movements. In a simple world, cobots need to perceive environments and quickly decide what path to choose. Cobots would enter any workspace and not struggle to provide flexibility.
Mitsubishi aims to create a collision-free environment with standard industry technologies but uses advanced algorithms to simplify dealing with unstructured and dynamic settings. Mitsubishi combined its efforts with Realtime Robotics, a unique company that has deployed transformative solutions such as risk-aware driving, high-production, and automated robotic vision for several automotive and industrial manufacturers. This collaboration is promising to accelerate results for obtaining a more flexible design in dynamic environments.
Realtime Robotics was able to layout the control and programming techniques that Mitsubishi Electric was missing. The solution is built around Realtime's motion planning accelerator hardware and RapidPlan Create software. RapidPlan removes the need for manual motions and robot interlocking, while the software dives into evaluating millions of alternative motion paths to avoid a collision before selecting an optimal path.
Key Take Away
Electrical engineers can find themselves working closely with the hardware side of robotics less than on the software side since that typically goes to computer engineers. The biggest takeaway for EEs is understanding that algorithms used to create human-aware robotics are the answer to innovating automation and industrial robotics.
Interested in other robotics news? Read more in the articles down below.