I am a graduate student (master degree) in Robotics and wants to keep pursuing my PHD degree in more Prestigious schools. Right now I am already participating in some research of Robot Arms. Which sub fields of Robot Arm research is the most promising ?

According my own experience, I have some recommendation to you as Robot Arm research:

1. Body direction. Simply put, it is to make a robotic arm. The domestic market is mainly concentrated in collaborative robotic arms. There was a wave of opportunities in 2019. Domestic collaborative robot companies such as Yuejiang got money and grew stronger at this time. The body direction is divided into joints and controllers. There are three types of joint technology. The first is qdd quasi-direct drive, which uses current loops to estimate force. Most of the domestically-produced robotic arms, including ur, are of this type. The second is sea, which is rarely used. The one that uses Rickson is of this type. The third is the torque sensor solution, which is used by Luoshi and Siling. For the controller, the domestic ones are basically the same. Linux plus real-time system patches are used as the basis, and then trajectory planning libraries, kinematic solutions, motion control master stations are added. The good ones also add dynamic models to the controller. The body direction is relatively volatile. Referring to Yuejiang Technology's IPO prospectus some time ago, it lost 100 million last year. Collaborative robots are not profitable at all. Half of the revenue is contributed by desktop robotic arms based on stepper motors. From a scientific research perspective, the body of the robotic arm is already very mature. Joints may be a good direction, but they are too low-level control, and their essence is motor control.

2. Grasping direction. Simply put, it is to buy a ready-made robotic arm and gripper, and then cooperate with vision, depth camera, etc. to do grasping applications. Robotic arm grasping is a very good research direction, which is divided into plane grasping and 6dof grasping. The representative algorithms of plane grasping are ggcnn and grcnn, and the representative algorithm of 6dof grasping is graspnet. I am optimistic about this direction, especially the new methods based on neural networks, reinforcement learning, and imitation learning. New algorithms emerge in an endless stream. It belongs to the strategy layer research of robotic arm control and is very valuable in scientific research.

3. Motion planning. Also known as path planning, simply put, it is to plan a path from the starting point to the end point in the robotic arm joint configuration space, avoiding all obstacles. The motion planning problem of the robotic arm is essentially the same as the path planning problem of mobile robots or drones in mathematics. However, the robot arm plans the algorithm in a configuration space with a better dimension. The traditional mobile robot method cannot be used to go to Star A. It is generally based on sampling methods such as rrt prm and a series of methods derived from it. Another type is to combine reinforcement learning. The scientific research value is also good, which belongs to the research of the robot arm strategy layer algorithm.

4. Force control. I am not doing force control research, so I will briefly talk about my understanding here. Force control includes impedance control and admittance control. The overall force control of the robot arm is studied based on joint force control or end force sensor. It was also a research hotspot before, but it has cooled down slightly in recent years. I personally think the core reason is that force control is not as important as imagined. In actual application scenarios, there are not many scenarios that require precise force control. In scientific research, it is a good direction, but it is also biased towards low-level control. I personally think it is not as good as the previous two.

5. Embodied intelligence direction. Embodied intelligence is a new term that came out this year, which is distinguished from pure software AI. In fact, it is to apply AI algorithms to robot control. For example, Aloha proposed a new imitation learning algorithm ACT and applied it to the Aloha device composed of a servo robot arm. This direction mainly focuses on the application of imitation learning algorithms and reinforcement learning algorithms on robot arms. I am very optimistic about this direction. I think this is the mainstream method of robot arm control in the future.

6. Go to the control direction and make a controller. If the teacher has money, he can help you apply for a set of robot arms and install them yourself, and then use ROS to make a set of control. That's about it.

7. Control theory and control engineering: This is a graduate direction directly related to automation. It mainly studies the control theory, methods and technologies of the system, including motion control of robots and robot arms, path planning, etc. In this field, you can study in depth how to achieve efficient and accurate control algorithms and apply them to actual robot systems.

8. Majors are not important. What matters most is the research direction of the tutor you choose. Robots are a hot topic, and many tutors have research directions in robots or robotic arms. If you want to make functional robots (soft robots, special robots, etc.), they may be more material-oriented, which I don’t think is very fun. If you make complex mechanisms and engage in hardware control, you must see whether the research group has sufficient funds and a strong team. If you run algorithms, do deep learning, vision, and simulation, the research group must at least have a high-performance workstation, and there must be more capable seniors to guide you to avoid pitfalls.

For more informtion, you can visit Robotic Search Engine to find latest research papers, AI Agent Marketplace and AI Agent Search Engine to find AI Agents related topics reviews and latest product information.